Simulation-Based Valuation of Project Finance Investments - Crucial Aspects of Power Plant Projects

TECHNISCHE UNIVERSITÄT MÜNCHEN
Lehrstuhl für Finanzmanagement und Kapitalmärkte
Simulation-based Valuation of Project Finance

Investments
– Crucial Aspects of Power Plant Projects
Dipl.-Kfm. Dipl.-Phys.
Matthäus Pietz
Vollständiger Abdruck der von der Fakultät für Wirtschaftswissenschaften der Technischen
Universität München zur Erlangung des akademischen Grades eines
Doktors der Wirtschaftswissenschaften

(Dr. rer. pol.)
genehmigten Dissertation.
Vorsitzender: Univ.-Prof. Dr. Christoph Ann
Prüfer der Dissertation: 1. Univ.-Prof. Dr. Christoph Kaserer
2. Univ.-Prof. Dr. Gunther Friedl
Die Dissertation wurde am 12.10.2010 bei der Technischen Universität München eingereicht und
durch die Fakultät für Wirtschaftswissenschaften am 15.12.2010 angenommen.
Contents
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
List of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
1 Introduction 1
1.1 Aims of Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Structure of Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Theoretical Framework 8
2.1 The Valuation of Project Finance . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Project Finance Fundamentals . . . . . . . . . . . . . . . . . . 10
2.1.1.1 Definition of Project Finance . . . . . . . . . . . . . . 10
2.1.1.2 Characteristics of Project Finance . . . . . . . . . . . 12
2.1.1.3 Motivation for the Use of Project Finance . . . . . . . 14
2.1.1.4 Project Finance Risks . . . . . . . . . . . . . . . . . . 15
2.1.1.5 Project Finance and Risk Management . . . . . . . . 17
2.1.1.6 Project Finance Market . . . . . . . . . . . . . . . . . 20
2.1.2 Capital Investment Valuation . . . . . . . . . . . . . . . . . . . 23
2.1.2.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . 23
2.1.2.2 Cost of Capital . . . . . . . . . . . . . . . . . . . . . . 25
2.1.2.3 Cash Flows . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.2.4 Cash Flow Modeling . . . . . . . . . . . . . . . . . . . 28
2.1.2.5 Stochastic Cash Flow Modeling . . . . . . . . . . . . 30
2.1.3 Project Finance Valuation . . . . . . . . . . . . . . . . . . . . . 34
2.1.3.1 Project Finance Specifics . . . . . . . . . . . . . . . . 34
i
2.1.3.2 Debt Perspective . . . . . . . . . . . . . . . . . . . . . 35
2.1.3.3 Equity Perspective . . . . . . . . . . . . . . . . . . . . 38
2.1.3.4 Related Literature . . . . . . . . . . . . . . . . . . . . 39
2.2 The German Electricity Wholesale Market . . . . . . . . . . . . . . . . 41
2.2.1 Liberalization of Electricity Markets . . . . . . . . . . . . . . . 42
2.2.1.1 Motivation and Market Reform . . . . . . . . . . . . . 42
2.2.1.2 Liberalization of the German Market . . . . . . . . . 43
2.2.1.3 Establishment of Electricity Exchanges . . . . . . . . 44
2.2.2 Stylized Facts on Electricity and on Electricity Prices . . . . . 46
2.2.2.1 Stylized Facts on Electricity . . . . . . . . . . . . . . 46
2.2.2.2 Stylized Facts on Electricity Prices . . . . . . . . . . . 48
2.2.3 German Electricity Market . . . . . . . . . . . . . . . . . . . . 50
2.2.3.1 Market Size . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2.3.2 Market Design . . . . . . . . . . . . . . . . . . . . . . 51
2.2.3.3 Market Structure . . . . . . . . . . . . . . . . . . . . 53
2.2.3.4 Retail Electricity Price . . . . . . . . . . . . . . . . . 54
2.2.4 European Energy Exchange . . . . . . . . . . . . . . . . . . . . 56
2.2.4.1 General Remarks . . . . . . . . . . . . . . . . . . . . . 56
2.2.4.2 Market Structure and Traded Products . . . . . . . . 58
2.2.4.3 Trading Process . . . . . . . . . . . . . . . . . . . . . 62
2.2.4.4 Market Participants . . . . . . . . . . . . . . . . . . . 67
2.2.4.5 Wholesale Electricity Price . . . . . . . . . . . . . . . 69
2.3 Price Formation in Commodity and Electricity Futures Markets . . . . 72
2.3.1 Introductory Remarks . . . . . . . . . . . . . . . . . . . . . . . 73
2.3.2 Price Formation in Commodity Futures Markets: Theory . . . 74
2.3.2.1 Theoretical Approaches . . . . . . . . . . . . . . . . . 74
2.3.2.2 Specifics of Electricity Futures Markets . . . . . . . . 78
2.3.3 Price Formation in Commodity Futures Markets: Evidence . . 80
2.3.3.1 Empirical Evidence Commodity Futures Markets . . . 80
2.3.3.2 Empirical Evidence Electricity Futures Markets . . . 82
3 A Project Finance Valuation Tool 93
3.1 Research Question . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.2 Functionality and Implementation . . . . . . . . . . . . . . . . . . . . 95
3.2.1 Introductory Remarks . . . . . . . . . . . . . . . . . . . . . . . 95
3.2.2 Functionality of the Tool . . . . . . . . . . . . . . . . . . . . . 96
3.2.3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3 The Valuation Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.3.1 Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.3.1.1 Input Parameters . . . . . . . . . . . . . . . . . . . . 99
3.3.1.2 Input Data . . . . . . . . . . . . . . . . . . . . . . . . 100
3.3.2 Computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.3.2.1 Cash Flow Equation . . . . . . . . . . . . . . . . . . . 101
3.3.2.2 Input Parameter Forecast . . . . . . . . . . . . . . . . 103
3.3.2.3 Monte Carlo Simulation . . . . . . . . . . . . . . . . . 104
3.3.2.4 NPV Calculation . . . . . . . . . . . . . . . . . . . . . 105
3.3.3 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.3.3.1 Output File . . . . . . . . . . . . . . . . . . . . . . . . 106
3.3.3.2 Output Data . . . . . . . . . . . . . . . . . . . . . . . 106
3.4 Status Quo and Limitations of the Valuation Tool . . . . . . . . . . . 107
3.4.1 Status Quo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.4.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.5 Excursus: Time Series Modeling and Forecasting . . . . . . . . . . . . 110
3.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.5.2 Forecast Models Level Data . . . . . . . . . . . . . . . . . . . . 111
3.5.3 Forecast Models Volatility . . . . . . . . . . . . . . . . . . . . . 115
3.5.4 Forecast Models Correlation . . . . . . . . . . . . . . . . . . . . 119
3.6 Concluding Remarks and Future Research . . . . . . . . . . . . . . . . 121
4 Price Formation in the German Electricity Wholesale Market – An

Empirical Analysis 123
4.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.4.1 Market Liquidity . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.4.1.1 Liquidity Spot Market . . . . . . . . . . . . . . . . . . 130
4.4.1.2 Liquidity Futures Market . . . . . . . . . . . . . . . . 133
4.4.2 Descriptive Results . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.4.2.1 Spot Market . . . . . . . . . . . . . . . . . . . . . . . 135
4.4.2.2 Futures Market . . . . . . . . . . . . . . . . . . . . . . 149
4.4.3 Risk Premia in the German Electricity Market . . . . . . . . . 153
4.4.3.1 Risk Premia in Spot Contracts . . . . . . . . . . . . . 153
4.4.3.2 Risk Premia in Futures Contracts . . . . . . . . . . . 165
5 The Impact of Model Complexity on the Simulation Results 178

5.2 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
5.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
5.2.2 Base Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
5.3 Case Study: Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
5.3.1 Impact of Simulation Complexity . . . . . . . . . . . . . . . . . 183
5.3.2 Impact of Forecast Complexity . . . . . . . . . . . . . . . . . . 188
6 Summary and Conclusion 193
A Hourly Prices Intraday Market 197
B Hourly Prices Day-Ahead Market 201
Bibliography 205
List of Figures
2.1 Loan-based Financing of Projects versus Project Finance . . . . . . . . 11

2.2 Project Finance Risks . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 A Typical Project Finance Structure . . . . . . . . . . . . . . . . . . . 20
2.4 Global Project Finance Loans Volume 2003-2009 . . . . . . . . . . . . 21
2.5 Global Project Finance Loans Volume by Sector in 2009 . . . . . . . . 22
2.6 Scenario Analysis versus Simulation Analysis . . . . . . . . . . . . . . 30
2.7 A Probability Distribution Obtained by Stochastic Cash Flow Modeling 32
2.8 Electricity Exchange Landscape in Europe in 2007 . . . . . . . . . . . 45
2.9 Evolution of German End Consumer Electricity Price 1991-2007 . . . 55
2.10 Market Structure EEX . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.11 Trading Phases Day-Ahead Market EEX . . . . . . . . . . . . . . . . . 63
2.12 Market Participants EEX by Number and Country of Origin in 2009 . 67
2.13 Evolution of German Wholesale Electricity Price 2002-2010 . . . . . . 69
3.1 Schematic Functionality Project Finance Valuation Tool . . . . . . . . 97

3.2 Underlying Cash Flow Equation . . . . . . . . . . . . . . . . . . . . . 102
3.3 Output File – Cash Flow Distribution with Measures . . . . . . . . . . 108
4.1 Traded Volume Spot Market 2003-2009 . . . . . . . . . . . . . . . . . 131

4.2 Daily Number Contracts without Trading in the Intraday Market . . . 132
4.3 Traded Volume Futures and Forward Market 2003-2009 . . . . . . . . 133
4.4 Number Traded Month Futures with Respect to Time-to-Delivery . . 134
4.5 Daily Prices Day-Ahead Market . . . . . . . . . . . . . . . . . . . . . . 137
4.6 Daily Absolute Returns Day-Ahead Market . . . . . . . . . . . . . . . 137
4.7 Daily Prices on a Weekly Scale Day-Ahead Market . . . . . . . . . . . 139
v
4.8 Hourly Prices on a Daily Scale Day-Ahead Market . . . . . . . . . . . 141
4.9 Hourly Demand in the German Electricity Market . . . . . . . . . . . 142
4.10 Monthly Prices on a Yearly Scale Day-Ahead Market . . . . . . . . . . 144
4.11 Selected Hourly Price Time Series Day-Ahead Market . . . . . . . . . 148
4.12 Daily Prices Intraday Market . . . . . . . . . . . . . . . . . . . . . . . 149
4.13 Month Base Future and Underlying Spot Price . . . . . . . . . . . . . 150
4.14 Monthly Price Day-Ahead Market . . . . . . . . . . . . . . . . . . . . 153
4.15 Selected Risk Premia Time Series in Day-Ahead Market Contracts . . 158
4.16 Time Variation Daily Risk Premia Day-Ahead Market . . . . . . . . . 161
4.17 Relative Risk Premia in One Month Futures . . . . . . . . . . . . . . . 167
4.18 Risk Premia in Month Base Futures by Time-to-Delivery . . . . . . . 168
4.19 Relative Risk Premia in Month Base Futures by Time-to-Delivery . . 169
4.20 Relative Risk Premia in Month Peak Futures by Time-to-Delivery . . 169
4.21 Relative Risk Premia in Month Base Futures by Year . . . . . . . . . 171
5.1 Dimensions of Model Complexity . . . . . . . . . . . . . . . . . . . . . 180

5.2 Impact Number of Iterations on NPV . . . . . . . . . . . . . . . . . . 183
5.3 Impact Number of Iterations on Default Probability . . . . . . . . . . 184
5.4 Impact Number of Iterations on Computation Time . . . . . . . . . . 185
5.5 Impact Equity Valuation Method on NPV . . . . . . . . . . . . . . . . 186
5.6 Impact Time Resolution on NPV . . . . . . . . . . . . . . . . . . . . . 187
5.7 Impact Time Resolution on Default Probability . . . . . . . . . . . . . 187
5.8 Impact Volatility Forecast Model on NPV . . . . . . . . . . . . . . . . 188
5.9 Impact Volatility Forecast Model on Default Probability . . . . . . . . 189
5.10 Impact Correlation Forecast Model on NPV . . . . . . . . . . . . . . . 190
5.11 Impact Correlation Forecast Model on Default Probability . . . . . . . 191
A.1 Time Series Intraday Market, Hour Contract 1 - 4 . . . . . . . . . . . 197

A.4 Time Series Intraday Market, Hour Contract 13 - 16 . . . . . . . . . . 199
B.1 Time Series Intraday Market, Hour Contract 1 - 4 . . . . . . . . . . . 201

B.4 Time Series Intraday Market, Hour Contract 13 - 16 . . . . . . . . . . 203
List of Tables
3.1 Summary Input Parameters . . . . . . . . . . . . . . . . . . . . . . . . 101
4.1 Summary Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

4.2 Percentage Days without Trading in Month Futures by Year . . . . . . 136
4.3 Descriptive Statistics for Hourly Day-Ahead Market Prices (Working
Days) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.4 Descriptive Statistics for Hourly Day-Ahead Market Prices (Non-Working
Days) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.5 Descriptive Statistics for Daily Day-Ahead Market Prices by Year . . . 143
4.6 Descriptive Statistics for Block Contract Market Prices . . . . . . . . . 145
4.7 Descriptive Statistics for Hourly Intraday Market Prices (Working Days)146
4.8 Descriptive Statistics for Hourly Intraday Market Prices (Non-Working
Days) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
4.9 Descriptive Statistics Month Base Futures . . . . . . . . . . . . . . . . 151
4.10 Descriptive Statistics Month Peak Futures . . . . . . . . . . . . . . . . 152
4.11 Risk Premia in Block Contracts . . . . . . . . . . . . . . . . . . . . . . 154
4.12 Risk Premia in Day-Ahead Market Contracts (Working Days) . . . . . 156
4.13 Risk Premia in Day-Ahead Market Contracts (Non-Working Days) . . 157
4.14 Risk Premia in Day-Ahead Market Contracts by Delivery Month (Work-
ing Days) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
4.15 Risk Premia in Day-Ahead Market Contracts by Delivery Month (Non-
Working Days) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.16 Term Structure of Risk Premia . . . . . . . . . . . . . . . . . . . . . . 163
4.17 Risk Premia in Month Base Futures . . . . . . . . . . . . . . . . . . . 165
4.18 Risk Premia in Month Peak Futures . . . . . . . . . . . . . . . . . . . 166
viii
4.19 Risk Premia in Month Base Futures by Delivery Period . . . . . . . . 172
4.20 Risk Premia in Month Peak Futures by Delivery Period . . . . . . . . 173
4.21 Regression Risk Premia on Variance and Skewness of Underlying Spot
Price . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
List of Abbreviations
APT Arbitrage Pricing Theory
AR Autoregressive
ARCH Autoregressive Conditional Heteroskedasticity
ARMA Autoregressive Moving Average
BaFin Bundesanstalt für Finanzdienstleistungsaufsicht
BMWi Bundesministerium für Wirtschaft und Technologie
CAPM Capital Asset Pricing Model
CCC Constant Conditional Correlation
CF Cash Flow from Operations
CfD Contract for Difference
CFTC Commodity Futures Trading Commission
CO2 Carbon Dioxide
DCC Dynamic Conditional Correlation
DCF Discounted Cash Flow
DRA Debt Repayment Amount
DSCR Debt Service Cover Ratio
EAU Emission Allowance Unit
EBIT Earnings before Interest and Taxes
EBITDA Earnings before Interest, Taxes, Depreciation, and Amortization
ECC European Commodity Clearing
ECDP Expected Cumulative Default Probability
EEG Erneuerbare Energien Gesetz
EEX European Energy Exchange
x
EGARCH Exponential General Autoregressive Conditional Heteroskedastic-
ity
EMEA Europe, Middle East and Africa
ENTSOE European Network of Transmission System Operators for Elec-

tricity
EnWG Energiewirtschaftsgesetz
EU ETS European Union Emission Trading System
EUREX European Exchange
EXAA Energy Exchange Austria
FCFE Free Cash Flow to Equity
FCFF Free Cash Flow to the Firm
FFO Funds from Operations
FOCF Free Operating Cash Flow
GARCH General Autoregressive Conditional Heteroskedasticity
GDP Gross Domestic Product
GJR-GARCH Glosten, Jagannathan and Runkle - General Autoregressive Con-

ditional Heteroskedasticity
GW Gigawatt
IEA International Energy Agency
IFRS International Financial Reporting Standards
IGSSE International Graduate School of Science and Engineering
IRR Internal Rate of Return
kWh Kilowatt Hour
KWK Kraft-Wärme-Kopplung
LLCR Loan Life Cover Ratio
LPX Leipzig Power Exchange
MA Moving Average
MATLAB Matrix Laboratory
MW Megawatt
MWh Megawatt Hour

NPV Net Present Value
NPVCR Net Present Value Cover Ratio
NYISO New York Independent System Operator
NYMEX New York Mercantile Exchange
OTC Over-the-Counter
PFVT Project Finance Valuation Tool
PHELIX Physical Electricity Index
PJM Pennsylvania, New Jersey, and Maryland
PLCR Project Life Cover Ratio
PTR Physical Transmission Right
QMV Quasi-Market Valuation
SPC Special Purpose Company
SPE Special Purpose Entity
SPV Special Purpose Vehicle
TSO Transmission System Operator
TUM Technische Universität München
TWh Terawatt Hour
WACC Weighted Average Cost of Capital
WpHG Wertpapierhandelsgesetz
XETRA Exchange Electronic Trading

Chapter 1
Introduction
According to the International Energy Agency (IEA), “the capital required to meet
projected energy demand through to 2030 is huge, amounting in cumulative terms
to $26 trillion (in year-2008 dollars) – equal to $1.1 trillion (or 1.4% of global GDP)
per year on average ...”.1 Considering also the importance to economic development
of secure energy supply, ensuring sufficient capital investments in the energy sector
is and will remain one of the major challenges of this century.
One of the main tasks during this process is the construction of power plants, as
growth in electricity demand is projected at 76% from 2007 to 2030, translating
into a required capacity addition of approximately 4,800 GW. Consequently, around
50% of the total investments are required by the power sector. These numbers can
even be considered conservative estimates, based on at least two developments with
highly uncertain future implications. First, 1.5 billion people worldwide currently
have no access to electricity. Thus, it can be assumed that economic progress in
developing countries will result in increasing investments in power plants. Second,
and most important, changes in most countries’ electricity policies seem necessary in
order to dampen the effects of climate change. The construction of power generating
facilities based on renewable energies, designed to replace fossil-fired power plants,
is currently the preferred option in achieving this goal.2
Power plant ventures are characterized by long project lifetimes and high capital re-
quirements, thus being highly dependent on secure economic and legal environments.
However, the liberalization of the electricity markets in the 1990s transformed a
1
IEA (2009), p. 2.
2
Cf. IEA (2009) for a summary of the key facts regarding the future energy policy, investments,
and trends.
1
2
former stable and regulated sector into a partially competitive market, with focus
shifting over the years towards price and volume uncertainty, market liquidity, and
price volatility. Adjusting investment plans to these conditions poses significant dif-
ficulties to market participants. For the German electricity market, the situation is
perhaps even more difficult, as the market is also subject to political interventions
regarding the use of renewable energies.3 In addition, one can assume that the pre-
vailing uncertainty regarding the future of German nuclear power plants has been
affecting long-term decisions in the electricity sector for years.
Today, a significant part of new power plants is financed via project finance, which
is currently the preferred approach to financing large-scale investments.4 The main
characteristics of this financing method are risk sharing and a strong focus on the
expected cash flows, due to the non-recourse character of the debt capital. These
characteristics fit very well with power plant ventures, thus it can be assumed that
the increased uncertainty in recent years will solidify the use of project finance in
the power sector. Other potential drivers of this development include the significant
energy investments needed in developing countries, where project finance is based
on its legal structure often the only available financing method.
A correct valuation of a large capital investment financed via project finance depends
to a significant degree on the accurate forecast of the expected cash flows. In contrast
to other financing methods, where the debt capital is often collateralized, both the
equity provider and the debt provider focus on this cash flow forecast. By nature, the
valuation results are sensitive to the underlying assumptions; however, no consensus
on the appropriate forecast technique exists, with various modeling approaches being
applied in both research and real-life applications. This results in a multitude of best
practice approaches and parameter specifications, but also in a lack of understanding
of the crucial aspects of the valuation process. Furthermore, it is not clear whether
the valuation of individual projects, such as power plant ventures, is characterized
by specific aspects.
Based on these considerations, an understanding of how the valuation of a power
plant venture has to take place, both from a theoretical and an implementation
perspective, is of high importance. Assuming that the venture is financed via project
finance, the focus must be on the cash flow modeling. A through understanding of
the valuation process would permit the identification of its main drivers and also
allow the evaluation of the results. Moreover, it would allow the quantification of
the relationship between their robustness and modeling complexity.
3
Cf. section 2.2.3.3 for a discussion of these interventions.
4
Cf. for example Esty (2004).
3
1.1 Aims of Analysis
The current focus of research on the topic of project finance is primarily on its
qualitative aspects. Questions regarding the defining features, the determinants of
its use, the legal structure, and the contractual arrangement are broadly discussed.
On the other hand, the literature on project finance valuation is sparse, with only
a few papers investigating the specifics valuating capital investments financed via
project finance. Based on this insight, the goal of this dissertation is the development
of a valuation model for project finance. As this dissertation is part of a broader
research program regarding the future of energy supply, the focus of the valuation
model is on power plant ventures financed via project finance.
The theoretical accuracy of the valuation process, an understanding of the crucial
aspects of the valuation process, and the development of an effective, computer-based
tool are intended. Three questions are of interest and need to be answered to develop
such a valuation model.
The first question relates to the specifications for the development of a practicable,
computer-based tool for the quantitative valuation of power plant ventures financed
via project finance. I intend, together with two other doctorate students, to develop
and implement a valuation tool that is based on stochastic cash flow modeling, as
this is a sophisticated modeling approach of future cash flows. Stochastic modeling
enables the computation of probability distributions instead of point estimates, thus
resulting in more meaningful and far reaching results. In particular, this modeling
approach allows the estimation of ex ante default probabilities. In addition, I intend
to implement advanced forecast models for level data, volatilities, and correlations.
The final result is aimed to be a self-contained, computer-based tool, which is sup-
posed to be suitable for both equity and debt providers. The applied forecast models
are selected from recent academic literature, but although stochastic cash flow mod-
eling is a standard application in the financial literature, combining the individual
forecast models with the stochastic modeling to an efficient and applicable tool is
challenging. Furthermore, I intend to implement correlation structures and non-
analytical distributions as an extension in order to increase the theoretical accuracy
of the valuation results.
The second question relates to the choice of the electricity price time series, i.e. the
market segment of the wholesale market, for the calibration of the valuation model.
As the generated electricity is the only output in the case of a power plant venture,
it is assumed that the electricity price is the critical profitability parameter. When
examining the academic literature on electricity prices and on electricity markets,
it stands out that these markets and the observed electricity prices exhibit unique
4
characteristics. To some extent these characteristics are due to the multitude of

significant turnovers which the electricity markets have undergone over the last two
decades, starting with the liberalization in the 1990s. Thus, the handling of the now
exchange-traded commodity electricity is difficult. Furthermore, the characteristics
of the national electricity markets differ significantly, due to different generation tech-
nologies, a lack of satisfactory transmission capacities between the single markets,
and varying market structures. The focus of this dissertation is on the valuation
of a power plant venture located in Germany, resulting in a focus on the German
electricity market and prices. One of the unique characteristics of electricity is its
non-storability, at least from an economic point of view. This results in the identifica-
tion of every electricity contract as a forward contract, and thus, in a corresponding
focus on the analysis of the German electricity futures market. The result of non-
storability is a loose relation between the spot and the futures market; the theory
of storage as the standard theory of price formation in commodity markets is not
applicable in this case. Academic literature suggests that the risk premia approach is
an appropriate price formation mechanism in electricity futures markets. According
to this approach, electricity futures prices are built on the expected spot price at
maturity of the future and of a risk premium, with empirical literature supporting
this view for various markets. However, academic research on the German electricity
market is just evolving. Therefore, I intend to conduct an empirical analysis of the
German electricity market to contribute to this research. I aim to investigate the
reliability and suitability of the risk premia approach as a theoretical price formation
mechanism. I apply the ex post approach, i.e. risk premia are estimated as the price
difference between futures prices and realized spot prices, and analyze all market
segments that are or were in existence since the foundation of the German wholesale
market.
Finally, the third question relates to the impact of model complexity on the valuation
results. I define model complexity based on two dimensions, i.e. the simulation and
the forecasting complexity; two dimensions that are assumed to have a significant
impact on the valuation results. Simulation complexity refers to parameters of the
simulation process; forecasting complexity refers to the applied forecast models. It
is assumed that model complexity corresponds to valuation accuracy. However,
simplifying assumptions are normally preferred as they lead to the usability of a
valuation tool and an increased comprehensibility of the valuation results. I aim
to quantify the impact of certain aspects; an identification of the crucial aspects is
intended. Furthermore, the quantification of the impact of model complexity on the
valuation results should provide information on the appropriate balance of modeling
complexity and simplifying assumptions.
5
1.2 Structure of Analysis
This dissertation consists of a theoretical and empirical section, divided across a

total of six chapters.
After the introduction in Chapter 1, the theoretical framework of the dissertation is
laid out in Chapter 2. This chapter consists of three sections that cover the basics
necessary as background for the empirical section.
The first section deals with project finance and the valuation of project finance invest-
ments. First, a definition of project finance and a distinction between this financing
method and the traditional corporate finance is provided, followed by a discussion of
the characteristics and the risks of project finance. Considerable attention is devoted
to the motivation for using this financing method. A market overview is given and
the key sectors in which project finance is used are pointed out. Then, the funda-
mentals of capital investment valuation are presented. The focus is on the discounted
cash flow analysis with both the specifics of cash flow modeling and the cost of cap-
ital being addressed. Furthermore, stochastic cash flow modeling as a sophisticated
modeling approach is introduced. The fundamentals of project finance valuation are
addressed in the last part of the section, distinguishing between the equity and the
debt provider perspective and explaining the specific requirements of these investors.
The common profitability and risk measures, and their use in practice are explained,
followed by an review of the quantitative literature on project finance valuation.
The second section provides an overview on the German electricity wholesale mar-
ket. Starting with a description of the liberalization process of electricity markets, I
discuss the establishment of electricity wholesale markets as one of the main results
of this process. I subsequently report stylized facts on electricity as a commodity
and on electricity prices. Afterwards, the German electricity market is addressed,
with the focus on the market size, the structure of the supply and the demand side
as well as the recent development of the electricity price. The end of the section
provides an overview on the German wholesale electricity market, which is located
on the European Energy Exchange (EEX), the German electricity exchange. Consid-
erable attention is devoted to the market structure, the traded products, the trading
process, and the market participants. This is followed by a discussion on the evolu-
tion of the observed electricity prices since the introduction of exchange trading and
potential drivers of this evolution.
The last section of the theoretical framework provides an overview on price formation
in commodity futures markets and, in particular, in electricity futures markets. At
the beginning of the section, the theory of storage and the risk premia approach as the
two standard price formation theories are discussed. Afterwards, the suitability of the
6
risk premia approach as a price formation mechanism in electricity futures markets

is determined. In addition, the recent empirical literature is reviewed, starting with
a discussion on the empirical evidence on price formation in commodity futures
markets, followed by a detailed literature review on price formation in electricity
markets. In this context, I distinguish between empirical evidence on short-term
and long-term futures contracts.
The empirical section of the dissertation consists of three chapters. Chapter 3 intro-
duces the newly developed stochastic project finance valuation tool while Chapter 4
contains the empirical analysis of price formation in the German electricity wholesale
market. Chapter 5 addresses the impact of model complexity on the results obtained
using the valuation tool.
A description of the stochastic project finance valuation tool developed in this dis-
sertation is given in Chapter 3. The chapter begins with a discussion of the research
question and introductory remarks on the valuation tool, followed by a presentation
of its basic functionality, its implementation, as well as corresponding problems. Af-
terwards, a fundamental introduction of the valuation tool is provided, distinguishing
between three basic process steps of the valuation process – input, computation, and
output – in order to better illustrate the valuation process. I discuss the necessary
inputs, the logic of the valuation process, and the computed results while also point-
ing out the status quo and present limitations of the valuation tool. The last section
contains an excursus regarding time series modeling and forecasting where the fore-
cast models implemented within the valuation tool are introduced. The chapter ends
with a discussion of the obtained results and potential avenues for future research.
Chapter 4 contains the empirical analysis of the German electricity wholesale mar-
ket, starting with the research question and a discussion of the analyzed data. This
section also addresses the rationale behind the data selection based on a discussion
of market liquidity. Then, the applied methodology and potential problems are dis-
cussed before the empirical results for the spot and futures market are reported.
Furthermore, descriptive statistics on the price data are reported, followed by em-
pirical results regarding the risk premia. The results on the magnitude and the sign
of the risk premia and their potential term structure are summarized. I also present
evidence relating time-variation and seasonality of the risk premia. The chapter ends
with a discussion of the obtained results and potential avenues for future research.
The analysis of the impact of model complexity on the valuation results is presented
in Chapter 5, which starts with a discussion of the research question and an in-
troduction to the fundamentals of the case study. The next section illustrates the
parameters and the underlying relations between them, and subsequently defines a
base case parameter constellation. An understanding of the basic assumptions is
7
important as the focus is on the relative changes instead of the absolute results in
the case study. Results relating to the impact of model complexity on the valua-
tion results are reported in two steps. First, the impact of simulation complexity is
addressed, and second, the impact of forecast complexity is examined. The chap-
ter ends with a discussion of the obtained results and potential avenues for future
research.
Chapter 6 concludes the dissertation, also addressing the implications of the obtained
results and discussing concrete suggestions for further research.
Chapter 2
Theoretical Framework
The theoretical framework of this dissertation is discussed in three sections. The

first section deals with the valuation of project finance, the second with the German
electricity wholesale market, and the third section with price formation in commodity
futures markets.
The section on the valuation of project finance, i.e. the valuation of capital invest-
ments financed via project finance, serves as an introduction to the financing method
project finance, to the fundamentals on capital investment valuation, and to project
finance valuation. An understanding of this financing method and its specifics is
necessary for the newly developed project finance valuation tool in chapter 3 and its
application on power plants financed via project finance in chapter 5.
The section on the German electricity wholesale market covers the fundamentals
of the recent liberalization of electricity markets, the German electricity market,
and the German electricity exchange. Moreover, the unique characteristics of the
commodity electricity and of electricity prices are addressed. An understanding
of these fundamentals serves as a basis for the empirical analysis of the German
electricity market in chapter 4; the fundamentals are provided as the research on
electricity markets is growing but these topics are not covered by standard literature
so far.
The section on price formation in commodity futures markets deals with the theory
and empirical evidence on price formation in these markets, in particular electric-
ity futures markets. As the empirical analysis in chapter 4 focuses on the question
whether there is evidence for the risk premia approach being an appropriate theoret-
ical price formation mechanism in the German electricity futures market, a profound
understanding of the academic literature on this topic is necessary.
8
9
2.1 The Valuation of Project Finance
Project finance has gained remarkable popularity over the past decades and today en-
joys the status of the preferred financing method for large-scale investment projects.
A focus on stand-alone projects, the non-recourse character of the provided debt
capital, a strong dependence on an accurate estimation of the future cash flows, and
an extensive risk-sharing are the main characteristics of this financing method.
In this section I introduce the financing method project finance. I begin with a
definition of project finance and a discussion of its characteristics and the rationale
behind its use. Moreover, I address the differences of this financing method to the
traditional corporate finance. Devoting considerable attention to the involved risks,
I then highlight the importance of risk management within a project finance invest-
ment, i.e. a capital investment financed via project finance. In addition, I provide
an overview of today’s market size and its astonishing growth over the last decades.
Afterwards, I focus on the valuation of capital investment projects. I present the
fundamentals with a separate discussion of the cost of capital of a company and of a
project; the modeling of the expected future cash flows is addressed. A discussion of
investment valuation based on cash flow modeling and its specifics is then followed
by a demonstration of the benefits of the application of stochastic cash flow modeling
for this purpose. Thereafter, I address the valuation of project finance investments
and the different interests of the debt and the equity side. A review of the related
literature regarding the valuation of project finance ends this chapter.
This section aims to answer the following three key questions:
• What is project finance and what are the characteristics, benefits, as well as
risks of this popular financing method?
• What is the appropriate method to valuate capital investments and what are
the specifics that have to be taken into account in the case of project finance?
• Is stochastic cash flow modeling an appropriate approach in project finance

valuation and what are its advantages compared to other modeling techniques?
10
2.1.1 Project Finance Fundamentals
2.1.1.1 Definition of Project Finance
Finnerty (2007) defines project finance as “the raising of funds on a limited-recourse

or non-recourse basis to finance an economically separable capital investment project
in which the providers of the funds look primarily to the cash flow from the project
as the source of funds to service their loans and provide the return of and a return
on their equity invested in the project”.1 Similarly Nevitt & Fabozzi (2000) classify
project finance as “a financing of a particular economic unit in which a lender is
satisfied to look initially to the cash flows and earnings of that economic unit as the
source of funds from which a loan will be repaid and to the assets of the economic
unit as collateral for the loan”.2 Finally, Esty (2004) sees project finance as “the
creation of a legally independent project company financed with equity from one or
more sponsoring firms and non-recourse debt for the purpose of investing in a capital
asset”.3
There is one key element of project finance, expressed in these definitions, that is
essential for this dissertation: Debt and equity providers are primarily dependent
on the project’s cash flow. This dependence on the expected cash flow is seen as
one of the defining features of the financing method project finance in its current
form.4 While this is a common case for equity providers, debt providers often have
their loans collateralized. Thus, the risk of a debt provider involved in a project
finance investment is significantly higher compared to a traditional corporate finance
investment.
The legal cornerstone of project finance is the creation of an independent project
company, generally called a Special Purpose Vehicle5 (SPV). The SPV is responsible
for the realization of the economically separate project and it is stocked with both
equity and debt. The equity is provided by equity investors, generally called sponsors,
and the debt mostly by a syndicate of financial institutions.6 After the completion
of the project the SPV is usually dissolved. The SPV is created by the sponsors who
initiate the project and also negotiate the debt contracts with the bank syndicate.
The projects that are financed via project finance are mostly independent, i.e. stand-
1
Finnerty (2007), p. 1.
2
Nevitt & Fabozzi (2000), p. 1.
3
Esty (2004), p. 213.
4
The definition of project finance in its current form goes back to the Financial Accounting
Standard No. 47 from 1981; cf. Tytko (2003), p. 13.
5
Alternative terms for SPV are Special Purpose Entity (SPE) or Special Purpose Company
(SPC).
6
Cf. for example Simons (1993), Dennis & Mullineaux (2000), Pichler & Wilhelm (2001), and
Sufi (2007) for a discussion of the rationales of bank syndicates.
11
Figure 2.1
Loan-based Financing of Projects versus Project Finance
Loan-based Financing Project Finance

of Projects
Bank(s) Bank(s)
Company SPV Company
Project Project Contractual Relationship
Source: Own work.
alone projects. Furthermore, the projects are often very specific, characterized by
low redeployability7 and a limited economic lifetime.
The creation of an independent company, in the form of the SPV, allows the sponsors
to limit their equity involvement to the initial capital amount. The SPV is the sole
borrower and all rights and responsibilities within the project are taken by this entity.
In the case of a project default, the SPV is the only entity filing for bankruptcy. Thus,
the sponsors primarily have the obligation to contribute their equity stake. Normally,
there is no collateralization of the debt capital on the part of the sponsors and the
debt is non-recourse.8
In the following it is important to distinguish between the terms project finance and
project financing. In this dissertation, project financing will be used as a general
term relating to the financing of a project without specifying how the project is
financed; project finance, on the other side, will be used for the specific financing
method that is defined above.
Figure 2.1 illustrates the general structure of a project finance investment versus a
7
Cf. Habib & Johnsen (1999), pp. 707-708.
8
However, contractual agreements with collateralization features also exist; these contracts are
known as limited recourse.
12
traditional corporate finance, i.e. a loan-based financing of a project.9 The main

difference between the loan-based financing and the project finance is that, in the
first case, the company acts as borrower and the debt capital is secured with its whole
balance sheet.10 If the project fails, the company may be forced into bankruptcy if
it is not able to repay the debt. In the case of a project finance investment, only the
SPV has legal obligations towards the bank(s). The maximum amount the sponsors
stand to lose is capped at the level of its total equity stake. This difference in
liability has significant implications regarding the granting of debt capital, which
will be discussed in section 2.1.3.2.
2.1.1.2 Characteristics of Project Finance
The following characteristics are, in general, attributed to the financing method

project finance:
1. Cash flow-related lending,
2. Non-recourse lending,
3. Off-balance-sheet financing,
4. Risk-sharing, and
5. High leverage.
The first characteristic – project finance as a cash flow-related lending method –

is based on the legal structure of project finance. Since the SPV acts as the sole
borrower – and because the SPV is, in general, exclusively founded for the financing
of the project – only two sources of collateral are available. The first source is
the assets acquired within the project; the second source is the expected cash flow.
In general, the acquired assets are very specific; low market values and liquidation
proceeds are the consequence.11 Thus, from a debt provider’s point of view, the
success of a project finance investment is almost solely dependent on the profitability
of the specific project, i.e. the ex ante unknown and uncertain cash flows.12
The second characteristic of project finance is that the provided debt capital is
normally non-recourse. This is due to the legal status of the SPV as a stand-alone
9
Cf. Ghersi & Sabal (2006) for a schematic comparison of corporate finance and project finance.
10
Cf. section 2.1.3.1 for a discussion of the traditional loan process.
11
Project finance is mainly used for large-scale investments; the capital assets acquired within
these transactions often have very specific designs and dimension and they are often not even trans-
portable.
12
Kleimeier & Megginson (2001) state that “project finance is used primarily to fund tangible-
asset-rich and capital intensive projects with relatively transparent (often hard-currency) cash flows”;
Kleimeier & Megginson (2001), p. 8.
13
entity. Non-recourse refers to the fact that the debt providers have no rights in
regards to the sponsors’ assets in case the project fails. This legal situation leads
to the debt providers taking on parts of the business risk13 . In order to partially
migrate this risk, debt providers have to estimate the expected cash flows of the
project over its entire lifetime as accurately as possible.14
The third characteristic – project finance as an off-balance-sheet financing – is also
closely related to the foundation of the SPV as a legally independent entity. Ac-
cording to the International Financial Reporting Standards (IFRS), shares below
50% in a SPV are to be recognized on the investor’s balance sheet using the eq-
uity method.15 Thus, it is only the equity stake, and not the whole balance sheet
of the SPV, that is activated when the share is below this threshold. In this case,
the sponsor avoids the decline of the equity ratio, due to the SPV’s often very high
debt-to-equity ratio as discussed below. Normally, a project finance investment is
initiated by one to three sponsors. However, in general, one majority shareholder
controls the project.16 Thus, the characteristic off-balance-sheet financing is not a
defining feature of project finance since it only holds for the minority sponsors.
Another common characteristic of project finance is risk-sharing. As will be more
fully discussed in section 2.1.1.4, in general, there are dozens of parties involved in
a typical project finance transaction. The rationale is that the involved parties try
to optimally allocate the individual risks to the parties that are able to bear them.
Therefore, packaging and transferring of risks is one of the main characteristics of this
financing method.17 Due to the resulting involvement of a high number of parties
and, hence, the high number of contracts that have to be negotiated Esty (2004)
remarks that “..., some people refer to project finance as ‘contract finance‘ ”.18
High leverage is another characteristic of project finance that is often discussed in
the corresponding literature. The high debt-to-equity ratios in project finance are
the direct result of the risk-sharing. An effective financing structure can be reached
through efficient risk-sharing, resulting in the ability to achieve a high ratio of debt
capital. Capital structures with high debt ratios are frequently observed; the aver-
age SPV has a debt-to-capital ratio of 70%, compared to an average ratio of 33% for
13
Cf. Kleimeier & Megginson (2001), p. 3.
14
Financial modeling, i.e. the application of advanced mathematical modeling techniques in the
context of finance, plays a more and more critical role in the context of project finance.
15
When applying the equity method to account for the investment in an associated company
the initial investment is activated at cost. Thereafter, the book value is increased or decreased to
recognize the investor’s share of the profits or losses; cash distributions received after the acquisition
decrease the book value. Cf. for example Penman (2007) for a further discussion.
16
Cf. Esty & Sesia (2010), p. 10. The authors also report that in 60% of all project finance
transactions only one sponsor is found on the equity side.
17
Cf. Marti & Keith (2000), p.2.
18
Esty (2004), p. 216.
14
public companies.19 The debt is generally provided by a bank syndicate. It is char-

acterized by larger absolute amounts and longer maturities versus other syndicated
loans.20
2.1.1.3 Motivation for the Use of Project Finance
When Airbus Industries decided to develop the A380 total project costs were es-
timated at 13 billion U.S. Dollar. For a company with yearly total sales of 17
billion U.S. Dollar this represents a capital amount that could obviously lead into
bankruptcy in the case of project failure. Therefore, the investment may be seen, as
stated by Esty (2004), as a kind of bet-the-company investment.21 It is thus not real-
istic to characterize the corporate manager facing this decision as the rational ‘accept
every positive net present value (NPV) project‘ decider. It rather seems traceable
that the management of a company might reject a positive NPV investment due to
concerns on the dimension and the risks. The fear is dragging the company into
bankruptcy in the case of project failure. Based on these considerations the aca-
demic literature postulates that the size of an investment project affects a manager’s
willingness to bear risk.22
Using project finance to finance large-scale projects has the advantage that the spon-
sor’s maximal loss is generally capped at its initial investment in the project. Thus,
managers might be willing to take more risk when using this financing method.
Financing of large-scale projects via project finance could have the aggregate result
that more profitable investments are conducted. Underinvestment due to managerial
risk aversion would also be reduced.23
Academic literature also suggests that project finance can reduce underinvestment
due to asymmetric information as introduced by Myers & Maljuf (1984) in their fa-
mous work on the determinants of capital structures. Furthermore, underinvestment
due to debt overhang could be reduced.24
In addition, the structure of project finance can diminish agency conflicts as intro-
duced by Jensen & Meckling (1976). According to Jensen’s Free Cash Flow Hypoth-
esis (Jensen (1986)) high debt-to-equity ratios serve as an important disciplinary
instrument: They prevent managers from wasting or mis-allocating free cash flows
19
Cf. Esty (2003), p. 7.
20
Esty & Megginson (2001) analyze project finance loan tranches. Using a sample of 495 project
finance loan tranches (with an overall worth of 151 billion U.S. Dollar) the authors find that the
largest single bank has on average a share of 20.3% on a single tranche and that the top five banks
hold 61.2% of a tranche.
21
Cf. Esty (2004), p. 220.
22
Cf. Esty (2004), p. 220.
23
Cf. Esty (2004), p. 217.
24
Cf. Esty (2003), p. 22.
15
to inefficient investments. Thus, the use of project finance reduces incentive conflicts
between the capital providers; agency costs are reduced and the expected free cash
flows increase as a result.25
Summarizing the arguments above project finance is another example for the inad-
equacy of the famous Proposition I – capital structure, respectively financing de-
cisions, do not matter in perfect markets – of Modigliani & Miller (1958) when
considering real, i.e. imperfect, markets. The financing of capital investments via
project finance can have the effect that sponsors undertake more risky projects than
they would be willing otherwise.26 Regarding capital investment in developing coun-
tries and emerging markets, for example, project finance plays a significant role. It is
often the only financing method available, in particular for large-scale projects. This
is due to the high risk of projects in these countries and markets; risks that a solid
project finance structure may face efficiently. Academic literature suggests that the
use of project finance in these countries serves as an important driver for economic
growth27 and helps filling the infrastructure gap.28
However, several drawbacks have to be mentioned, even though the popularity of
project finance seems to prove that market participants are willing to accept them.
One drawback is the additional time that it takes to create the SPV, taking from
6 up to 18 months longer compared to alternative financing methods. Another
disadvantage is represented by high transaction costs, reaching often 5-10% of the
project’s total costs.29 Also the up-front fees (plus advisory fees relating to the
structure of the project finance transaction) and interest rates are considerably higher
than for corporate finance transactions.30
2.1.1.4 Project Finance Risks
Project finance is exposed – similar to other financing methods – to a multitude of

risks. Before I discuss the individual risk, I first classify these risks in accordance with
the broad literature in four categories, namely technical, environmental, economic,
and political31 risks. Figure 2.2 contains a summary of the potential risks, classified
in the four categories, arising in a typical project finance investment.32
Technical risks associated with the construction of the project are at the main concern
25
Cf. Subramanian et al. (2007) for a theoretical model that allows to derive this result.
26
Cf. Culp & Forrester (2010), p. 2.
27
Cf. Kleimeier & Versteeg (2009), p. 3.
28
Cf. Hammami et al. (2006), p. 5.
29
Cf. Esty (2004), p. 216.
30
Cf. Subramanian et al. (2007), p. 1.
31
Cf. Hainz & Kleimeier (2006) for a discussion of the political risks in project finance.
32
Cf. also Decker (2008), pp. 145-148 and Przybilla (2008), pp. 93-100 for a compilation of
potential risks in project finance investments.
16
Figure 2.2
Project Finance Risks
Technical risks Environmental risks
• Risks caused by environmental effects

• Completion risk
on / of a project
• Latent defects risk
• Force majeure risk
• Risk of breakdown
(e.g earthquakes, floods, …)
• (Technical) underperformance
• Accidents
… …
Economic risks Political risks

• Commodity price risk
• Expropriation risk
• Cost overrun risk
• Currency convertibility risk
• Operational risk
• Transferability risk
• Currency risk
• Political violence risk (war, sabotage, terrorism)
• Market risk (Introduction of substitutes; better
• Regulatory risk
technology, …)
…
…
Source: Own work.
during the early stage of a project. In addition, the project is in particular exposed
to environmental risks in this stage. During the operational phase the main risks
associated with the underperformance or even failure of the project due to economical
or political circumstances become of main importance.33
Completion risk is the main technical risk. This risk includes the possibility that the
project might not be completed, often implying a total loss for both the equity and
debt providers. The failed completion of a project can be either cost- or technology-
related. Another technical risk is technical underperformance, meaning that a project
is completed but never reaches the projected output and hence not the expected level
of cash flows.
Environmental risks can be divided into two risk categories; generally they are
present for the whole lifetime of the project but of particular importance at the
beginning of the project. The first category reflects the effects of the project (con-
struction) on the environment (e.g. air pollution, ground water pollution, ...). These
effects can cause high costs and even force the sponsors to redesign the project. The
second risk category is related to the force majeure risks (e.g. fires, earthquakes,
...) and other not directly avoidable catastrophes that can seriously damage or even
33
Cf. Sorge (2004), p. 94.
17
completely destroy the project.

Following the technological project completion economic risk becomes the main con-
cern. Examples of economic risk are commodity price risk, i.e. higher input costs
than expected, and demand risk, i.e. the failure to reach the projected output levels
due to lower demand. Finally, political risks include risks such as the host govern-
ment changing the regulatory environment or devaluing the currency. Political risks
also include the risk of expropriation.
2.1.1.5 Project Finance and Risk Management
Addressing the risks discussed above, i.e. an effective risk management, plays a
distinct role in project finance. Although this is also for other financing methods
relevant, risk management is of special importance in project finance. This is due
to the fact that the expected cash flows represent the majority of collateral to the
debt providers. Thus, debt providers have a strong incentive to hedge as many risks
as possible in order to secure predictable cash flows. The aim of hedging is the
reduction of future cash flow variability and probability of default. It also reduces
potential difficulties of the SPV in meeting its regular interest and loan repayments
and thus the probability of a necessary restructuring. The existing risk factors need
to be hedged to an extent where future cash flows become predictable to acceptable
levels for the debt provider.
The debt providers’ requirement to hedge as many risk as possible confronts the
sponsors with the situation that the degree of risk management of a project not
only directly impacts the cost of debt but effectively also the accessibility to debt
capital. A certain minimum degree of hedging is expected by the market and has to
be provided. Thus, the sponsors do not face the question whether or not to hedge but
only to which extent. The corresponding literature indicates that as a consequence
hedging (at least through the non-financial contracts discussed below) takes place
before the negotiation of the debt contracts. This hedging activities of the sponsors
are even more amplified by the fact that the debt providers, almost exclusively banks,
have a favorable negotiating position since bank loans are the primary source of debt
capital in project finance.34
The possibilities to hedge the arising risks include mainly the use of insurance, of
financial contracts, and of non-financial contracts. Another possibility is the use
of third party guarantees which are particularly used in respect to certain political
risk.35
34
Cf. Beale et al. (2002), p. 6.
35
Cf. Sorge & Gadanecz (2008), p. 68.
18
Insurance is primarily used to hedge force majeure risks, e.g. fire, water damage or
terrorist attacks. Insuring force majeure risks is usually mandatory in project finance
investments. Also common is the insurance of political risks through export credit
agencies.36
Financial contracts are often used in project finance transactions for hedging short-
term risks. Examples include future contracts which are used to hedge the input and
output factor prices, not secured by long-term contracts. However, the vast majority
of financial contracts are used for hedging market-wide risks, i.e. macroeconomic
risks as foreign exchange risk, interest rate risk and inflation risk.
Non-financial contracts are the main hedging instrument used in project finance.
The sponsors usually sign a multitude of contracts to reduce the project’s risks,
i.e. to shift the risks from the SPV to third parties willing to bear them.37 The
result is a wide network of contracts, such as long-term agreements on the input and
output factors, precisely defined construction contracts and detailed operating and
maintenance contracts. Construction contracts, for example, are generally designed
in a manner that the responsible construction company is obligated to compensate
the SPV for delays and for other failures.38 In particular, it is important to hedge
the input and output factor price risk.39 Due to the popularity of project finance in
the energy and mining sector commodity price risk is widely seen as the main risk
in a project finance investment.40
Regarding the non-financial contracts Corielli et al. (2008) identify four types of
contracts that are critical for the soundness of a project finance investment:41
1. Construction contracts,
2. Purchasing agreements,
3. Selling agreements, and
4. Operation and maintenance agreements.
Construction contracts include all contracts concerning the completion of the project.
Purchasing and selling agreements include all contracts that intend to secure the
36
Cf. Klompjan & Wouters (2002), p. 2.
37
Cf. Lessard & Miller (2001), pp. 12-14.
38
Another possibility to hedge the completion risk from the point of view of the debt providers is
to ask for a sponsors’ guarantee for delay risk. This kind of bilateral agreement is also used for other
specific risks where the debt providers secure their positions through contracts with the sponsor.
39
Take-or-Pay contracts and Put-or-Pay contracts are examples for hedging instruments regarding
the input and output factor risks; cf. Tytko (2003), p. 32.
40
Cf. Spillers (1999), p. 1.
41
Cf. Corielli et al. (2008), pp. 9-10.
19
input and the output prices. Operating and maintenance agreements are contracts
that aim to secure the current costs.
Summarizing, project finance is characterized through a network of (non-financial)
contracts.42 The advantage of these contracts is that they can be used to address
risks that are project-specific and that cannot be hedged in financial markets.43
The aim is to reduce the cash flow variability and to avoid the negative impact of
unexpected events causing a decrease of the cash flows. Esty (2003) estimates that
a typical project finance transaction involves around 15 different parties and 40 or
more contracts. Larger deals can involve significantly higher numbers of parties and
contracts.44 Gatti et al. (2008) find that debt providers seem to rely on the various
contracts not just as a mechanism for controlling project risks but also agency costs,
an additional motivation for the use of project finance discussed in the section above.
Another risk reduction mechanism often found in project finance investments is the
use of covenants.45 As a consequence of the interest divergence between sponsors
and debt providers project finance is characterized by an intense use of covenants46
in the debt capital contracts.47 The debt providers are using covenants to align
the sponsors’ interest with their own, i.e. to improve the monitoring quality and to
reduce managerial discretion.48
The creation of reserve accounts is another risk reducing instrument used by debt
providers. A reserve account is created for the case that the cash flows of one or
more consecutive periods are not sufficient to serve the mandatory debt payments.
To avoid project failure due to a few periods with cash flows lower than assumed, the
sponsors usually agree not to withdraw dividends until a certain amount has been
deposited in specific reserve accounts.
Figure 2.3 shows a typical project finance structure as found in the energy industry.49
The project is a gas-fired power plant. The project is financed by two sponsors on the
equity side and a bank syndicate on the debt side. In this example the contractual
structure consists of a supply contract for the gas input and a purchase contract for
the power output. In reality these contracts often have long maturities, sometimes
42
Cf. Dailami & Hauswald (2001) for a discussion of project finance as a nexus of contracts.
43
Cf. Dailami et al. (1999), p. 2.
44
Cf. Esty (2003), p. 8.
45
Cf. Fahrholz (1998), pp. 276-278.
46
Covenants are legal clauses included in loan contracts. They specify one or more conditions
that have to be obligated by the borrowing party. In the case of a breach of the covenants the lender
often has the possibility to call the loan within a short term.
47
However, parts of the empirical literature find that project finance is characterized by a lower
number of covenants than in corporate finance; cf. for example Kleimeier & Megginson (2001).
48
Cf. for example Rajan & Winton (1995) or Röver (2003) for a discussion of the motivation for
implementing covenants in debt contracts.
49
However, the illustration is strongly simplified. For the discussion of a more realistic project
cf. for example Bonetti et al. (2010).
20
Figure 2.3
A Typical Project Finance Structure
Bank
Sponsor A Sponsor B
Syndicate
Non-recourse
Equity
Debt
Labor
Gas Input Power Output

under a supply SPV under a purchase
contract contract
Technol.
License
Equipment Construction Operating &

contract contract Maint.contract
Host Government:
legal system, permits, regulation,
property rights, etc.
Source: Modified, taken from Esty (2003), p. 36.
decades.50 In addition, a construction and equipment contract secures the initial

capital investment. Labor contracts and operating and maintenance contracts secure
the current costs. The legal environment necessary for the safe operation of the power
plant is provided by the host government.
2.1.1.6 Project Finance Market
The financing of a gas and oil project in the 1930s by a Dallas bank is seen by
some authors as the birth of the financing method project finance as it is known
today.51 However, financing structures with similar characteristics can be traced
back to approximately 700 years ago.52
The exploration of the North Sea oil fields in the 1970s marks the first use of project
finance on a larger scale.53 The financial success of this project established project fi-
50
The ability to hedge a project’s output is of course dependent on the sector in which the project
finance is conducted. Long-term hedging is common in the oil, gas, and power sector. However, it
is not possible in the transportation or hotel sector.
51
Cf. for example Culp & Forrester (2010), p. 2.
52
Cf. for example Finnerty (2007), p. 4; cf. also Esty & Sesia (2010) for an overview of the
history of project finance.
53
Cf. Kleimeier & Megginson (2001), p. 3.
21
Figure 2.4
Global Project Finance Loans Volume 2003-2009
Source: Taken from Thomson Reuters (2009), p. 1.
nance worldwide as the preferred financing method for large-scale investments. Esty
(2004) reports that today approximately 10 to 15% of capital investments are fi-
nanced via project finance in the United States. For investments with a volume
exceeding 500 million U.S. Dollar this ratio even reaches 50%.54
The worldwide volume of project finance has risen considerably over the last decades,
from less than 10 billion U.S. Dollar annually in the late 1980s to approximately 328
billion U.S. Dollar annually in 2006.55 During the 1990s the compound annual growth
rate has been almost 20%.56 Corielli et al. (2008) report that the compound annual
growth rates for project finance loans and project finance bonds between 1994 and
2006 were recorded at 23% and 15%, respectively.57 After an all time high of 408
billion U.S. Dollar in 2008 the market volume declined to 240 billion U.S. Dollar in
2009.58 This decline of over 40% reflects the worldwide financial and later economic
crisis.59 In the sections above it was indicated that the risks of a debt provider in
project finance are by far higher than in a corporate finance and in other financing
54
Cf. Esty (2004), p. 214.
55
Cf. Esty & Sesia (2007), p. 1.
56
Cf. Esty (2004), p. 213.
57
Cf. Corielli et al. (2008), p. 3.
58
59
In 2002, the last worldwide economic crisis, an estimated 40% decline in the project finance
market was observed; cf. Esty (2003), p. 1.
22
Figure 2.5
Global Project Finance Loans Volume by Sector in 2009
Source: Taken from Thomson Reuters (2009), p. 1.
methods. As banks tend to cut risky investments and loans in the first round of a
crisis, project finance loans were the first to be cancelled after the eruption of the
current crisis in 2008.
According to Hainz & Kleimeier (2004) a total amount of 963 billion U.S. Dollar
has been raised between 1980 and 2003 for project finance loans.60 Figure 2.4 shows
the development of the global project finance loans volume over the last seven years.
The total loan volume is broken down in the three regions Asia Pacific, Americas,
and Europe, Middle East and Africa (EMEA). Apparently the strong growth from
the 1990s continued into the new millennium with the market peaking in 2008.
Regarding the use of project finance by industries most applications are found in the
infrastructure sector (e.g. toll roads), energy sector (e.g. power plants), and natural
resource sector (e.g. mines).61 Figure 2.5 breaks down the 2009 project finance loans
volume across the sectors, showing that almost 40% of the whole market consists of
power projects. Oil and gas (18%) and Transportation (18%) are the other two
significant sectors in the market.
The most prominent examples of project financed investments in recent years include
the four billion U.S. Dollar Chad-Cameroon pipeline, the six billion U.S. Dollar
60
Cf. Hainz & Kleimeier (2004), p. 2.
61
Cf. Esty (2004), p. 214.
23
Irdium global satellite telecommunications system, the 18 billion U.S. Dollar Papua
New Guinea Liquified Natural Gas project and the 29 billion U.S. Dollar Sakhalin
II gas field.62
Looking back to the 1990s, the construction of the Disneyland Europe and of the Eu-
rotunnel between France and the United Kingdom are probably the most prominent
projects financed by project finance. In particular the Eurotunnel is an interesting
case study as it was the largest capital investment ever to be financed via project
finance. It ended in a financial disaster from which many lessons can be learned.63
The significant cost underestimation observed during the Eurotunnel construction is
also characteristic for many other projects.64
2.1.2 Capital Investment Valuation
2.1.2.1 Fundamentals
The term valuation, in general, refers to the determination of the economic value of
an asset, liability or capital investment.65 The valuation process is mostly based on
an approach such as the discounted cash flow (DCF) analysis or relative valuation
methods such as the multiples method.
In the following, I focus on the valuation in the context of capital budgeting, i.e. the
valuation of capital investments.66
The valuation of capital investments is focused on the question whether investing
in a certain project makes economic sense. The DCF analysis is, at least when
large capital investments (under certainty) are valuated, the most common concept
applied today.67 However, other methods as the internal rate of return (IRR) or the
payback period are used as well.68
The DCF analysis is based on the calculation of the present value of the future
cash flows, i.e. the future cash flows are estimated and discounted. The sum of all
discounted cash flows, both incoming and outgoing, is the NPV.69 The basic decision
62
63
Cf. Vilanova (2006) for a discussion of the Eurotunnel venture.
64
Cf. Flyvbjerg et al. (2002) for an analysis of the phenomenon cost underestimation in trans-
portation projects.
65
I always use the term value as market value in this dissertation.
66
The theory for the valuation of investment projects has significantly improved over the last
century. Taking the equity provider’s point of view, the payback method has been the most common
concept for determining profitability at the beginning of the last century. Nowadays, the DCF
analysis is the most common concept. Recent literature has already begun discussing whether the
real options approach is more suitable to value investment projects, at least in certain industries.
67
Cf. Graham & Harvey (2001) for a survey on the valuation methods used in practice.
68
Cf. for example Ross et al. (2005) for a discussion of alternative investment rules.
69
Cf. Fernandez (2009) for a discussion of the various valuation methods based on cash flows.
24
rule underlying the NPV method is going forward with projects that have a positive
NPV and rejecting projects with a negative NPV.70
Primarily, the DCF analysis is used to value equity investments in two ways: First,
indirectly by discounting the free cash flow to the firm (FCFF)71 with the help
of the cost of capital72 and subtracting the debt value (adjusted by cash reserves)
afterwards. Second, by directly discounting the free cash flow to equity (FCFE)73
using the cost of equity.
In the first case, the result of the FCFF approach is the entity value defined as
n
X F CF Ft
N P VEntity V alue = (2.1)
t=1
(1 + rc )n
with F CF Ft the free cash flow to the firm in period t, rc the cost of capital (assumed
to be constant) and n the lifetime of the investment (project, enterprise, etc.).
In the second case, based on the same assumptions as above, the result of the FCFE
approach is the equity value
n
X F CF Et
N P VEquity V alue = . (2.2)
t=1
(1 + re )n
Instead of the free cash flow to the firm the FCFE is used. In addition, the cost of
capital is replaced by the cost of equity, re .
The equity value reflects the value of an investment to the equity providers; the
entity value reflects the total value of the investment to all capital providers. In
reality, the FCFF approach is more common because, as stated by Copeland et al.
(2000), “discounting equity cash flows provides less information about the sources of
value creation and is not as useful for identifying value-creation opportunities”.74
Two types of issues arise at this point: First, the determination of the risk-adjusted
cost of capital (equity) is not straightforward. Second, it is of significant importance
to account for the uncertainty of the future cash flows.
The appropriate method to determine the cost of capital is discussed in the next
section, whereby the determination of the cost of equity is the crucial task in real-life
applications. Regarding the uncertainty of the future cash flows Vose (2000) points
70
Cf. for example Ross et al. (2005), Copeland et al. (2005), and Brealey et al. (2007) for a
discussion of the NPV method.
71
The FCFF is the amount of generated cash that can be – after deduction of the necessary
reinvestments – distributed to both the debt and the equity providers.
72
Cf. the next section for a definition of the cost of capital for a company.
73
The FCFE is defined as the FCFF minus debt repayments and changes in the net debt capital.
74
Copeland et al. (2000), p. 151.
25
out that the main solutions that allow for the consideration of the future uncertainty
in capital investments are either a deterministic scenario-based modeling approach or
the application of stochastic modeling. Both methods are discussed in the following
including their advantages and their drawbacks.
2.1.2.2 Cost of Capital
The cost of capital of a company is the average cost of a company’s funds.75 The
funds of a company normally consist of a mix of equity and debt capital. In addition,
capital categorized in between these two fundamental capital sources exists in reality
as well.76 The weighted average cost of capital (WACC) is, in general, used as
a measure of the average capital cost.77 The cost of capital of a company is the
expected return of a hypothetical investor allocating capital in a portfolio of all
the company’s existing securities.78 In reality, the WACC is mostly calculated as a
weighted average of the cost of equity, i.e. the expected return on the current equity
capital, and the cost of debt, i.e. the expected return on the current debt capital.
In the case of an capital investment the cost of capital is the minimum rate of return
that an investor expects from his investment. The same return can be earned by
investing in an alternative project with equivalent risk (opportunity costs). Hurdle
rate is another term used in this context.
Cost of equity
The cost of equity is usually estimated with the Capital Asset Pricing Model (CAPM),
extensions of the CAPM or with the Arbitrage Pricing Theory (APT).79 The CAPM
is by far the most often used approach in real-life applications.80
The CAPM states that the cost of equity is equal to the sum of the risk-free rate
and the equity risk premium times a factor related to the systematic risk, called beta
in the financial literature.81 Two points relating to the estimation of the systematic
75
Cf. Copeland et al. (2005) for a discussion of the cost of capital and their determination or,
rather, estimation.
76
Funds categorized between equity and debt are usually called mezzanine capital.
77
The WACC is defined as
Equity Debt
W ACC = re + (1 − T ) rd
Equity + Debt Equity + Debt
where re are the cost of equity and rD the cost of debt. T is the tax rate and Equity and Debt the
market values of the equity and debt capital, respectively.
78
Cf. Brealey et al. (2007), p. 215
79
The CAPM was developed by Sharpe (1964), Lintner (1965), and Mossin (1966). The APT was
developed by Ross (1976). Both models build on the work of Markowitz (1952) on diversification
and portfolio theory.
80
Cf. Graham & Harvey (2001), p. 201.
81
Cf. Fama & French (2004) for a discussion of the theory and empirical evidence regarding the
26
risk are of interest for this dissertation: First, what is the right beta to valuate a
capital investment and second, is this beta stable over the lifetime of the project or
is it rather time-dependent.
An application of the CAPM for the estimation of a company’s beta results in an
estimate of the cost of equity of the company. However, this cost of equity can only
be used for the valuation of a capital investment when the risk of the investment is
equal to the average risk of the company. In addition, the investment needs to be
financed with the same capital mix as the actual mix of the company. Otherwise, an
appropriate levered beta must be used.82 In the case that the risk of the investment
is not equal to the company’s risk a specific cost of equity for the investment must
be determined.83
Regarding the time-dependence of the beta, the DCF analysis in classical valuation
theory is mainly applied to industry companies with rather stable capital structures.
Therefore, a single, constant discount rate can be used. However, this procedure is
not appropriate for the valuation of capital investments financed via project finance
since they are characterized by limited lifetimes and time-varying capital structures.
One solution recommended by the academic literature is the assumption of a target
capital structure which allows the calculation of a discount rate.84 This approach
may lead to a miscalculation of the project’s cost of equity since the costs of equity
depend on the project’s capital structure.85 Damodaran (1994) and Grinblatt &
Titman (2001) argue that only the application of different discount rates for every
period based on the actual capital structure produces an unbiased cost of equity.86
Esty (1999) argues that it is crucial to calculate the cost of equity based on market
values of debt and equity instead of their book values. However, this leads to a
circularity problem: The market value of equity, which is calculated as the sum of
the book value of equity and the NPV, cannot be calculated since the cost of equity
is necessary for that. A possible solution, first applied to project finance valuation
by Esty (1999), is the so called quasi-market valuation (QMV).87 According to this
method the market value of equity at the end of the first year equals the initial book
value of equity plus the expected NPV of the project.
CAPM.
82
Cf. Kaplan & Ruback (1995), pp. 1064-1068.
83
Cf. Brigham et al. (1999), pp. 168-175.
84
Cf. for example Ehrhardt (1994) and Finnerty (2007).
85
Cf. Esty (1999) for a discussion of this problem. This mis-estimation of the cost of equity also
distorts the calculation of the WACC.
86
I use this procedure in the newly developed project finance valuation tool by determining the
cost of equity each time the cash flow is calculated. Cf. chapter 3 for a discussion of the exact
calculation procedure.
87
The QMV is based on three assumptions, namely that (i) the CAPM is valid, (ii) the market
value of debt is equal to the book value, and (iii) the market is efficient.
27
Due to the advantages of the QMV method, it is implemented and used in the newly
developed project finance valuation tool which is introduced in chapter 3.88
Cost of debt
The calculation or rather estimation of the cost of debt is mostly straightforward.
Common methods to determine the cost of debt are, for example, the risk premium
approach, i.e. the cost of debt are estimated as the sum of the risk-free rate and an
appropriate risk premium, or the benchmarking with comparable debt securities. I
do not discuss the individual approaches separately here and refer to the standard
financial literature. However, I note that it is important to distinguish whether the
cost of debt is reported before or after tax. Normally, they are reported before tax
but since the actual cost of debt is reduced by the level of the company’s tax benefit
the effective cost is equal to the cost of debt after tax.
In the case study in chapter 5, the cost of capital is seen as a constant input factor
which is specified at the beginning of the valuation process and remains unchanged
thereafter.
2.1.2.3 Cash Flows
A cash flow, to use a simple and fundamental definition, is the difference between the
cash that moves in and out of a certain entity during a certain period. In contrast to
profits that are based on accounting principles, cash flows are based on the observable
movements of cash. To determine cash flows both the direct and the indirect method
can be used. The direct method sums up observed cash outflows and inflows, the
indirect method starts with an accounting measure, normally the net income, and
corrects it with all non-cash charges.89
One of the main advantages of the use of cash flows is its role as an alternate measure
of profits. It is widely believed that accrual accounting concepts, due to the wide
choice of parameters, do not represent economic realities, at least in the short-term.
In other words, insiders can use various parameters of choice to manipulate financial
results in their favor. Besides, it is assumed that it is more difficult to manipulate
cash flows than profits. Thus, almost all modern valuation methods are based on
cash flow measures and not on accounting profits.
88
Cf. chapter 3 for a discussion of the implementation of the QMV method within the developed
project finance valuation tool.
89
Three sources of cash flows are usually distinguished in the financial literature and financial
reports: Cash flows from (i) operating activities, (ii) investment activities, and (iii) financing activ-
ities. Operating cash flows are generated from activities related to selling products, investing cash
flows by purchasing assets, and financing cash flows result from cash transactions with claimants.
Cf. for example Penman (2007) for a further discussion.
28
2.1.2.4 Cash Flow Modeling
Cash flow modeling is an integral part of the capital budgeting process. The aim is
to find a methodology that is suitable to forecast future cash flows and to use this
method in such a way that the obtained results are meaningful.90
The most common approach used in cash flow modeling is the fundamental analysis.
Fundamental analysis relies on historical data, e.g. financial reports and historical
time series, and aims to derive future values based on these data. In a simple forecast
model values from the current financial statement are used to predict future values.
A simple extrapolation of the last value, the use of a deterministic trend or the
application of an advanced time series model are examples for applicable methods.
Another possibility is estimating the future cash flow based on pro forma financial
statements. Regarding the application of cash flow modeling and forecasting in
reality, the fundamental method is mostly the standard method used in real-life
applications.91
The application of cash flow models based on a fundamental analysis has one im-
portant drawback: The obtained future cash flows are normally point estimates.
The calculated NPV based on these cash flows will then also be a point estimate,
providing no information on the uncertainty of the future cash flows and of the NPV.
To overcome this drawback risk analysis methods can be applied to gain insight
into the risks of a capital investment. From a theoretical point of view, the use of
risk analysis allows the testing of the robustness of the estimated cash flows. The
following methods are suggested by the financial literature for this task:
1. Sensitivity analysis,
2. Scenario analysis, and
3. Simulation analysis.
A sensitivity analysis measures the effect of changes in one of the input parameters
on the estimated future cash flows.92 The scenario analysis allows the simultaneous
manipulation of several of the input parameters and the quantification of the impact.
Thus each scenario results in one cash flow. A simulation analysis is based on a
random manipulation of all input parameters. As a result it yields a probability
distribution of the future cash flows.
90
Cf. Brealey et al. (2007) for a discussion of the potential problems in the determination of the
relevant cash flows of a project.
91
Cf. Kaka (1996) for a discussion of cash flow forecasting based on a simple cash flow model.
92
Cf. Hwee & Tiong (2002) for an application of a scenario analysis in the context of a computer
based cash flow forecast model.
29
The use of sensitivity analysis on all input parameters allows the identification of the
input parameters which have the most influence on the output. The calculation of
various scenarios within the scenario analysis allows the estimation of the impact of
certain input parameter combinations on the output. Mostly three scenarios – called
a base case, a worst case, and a best case – based on input parameters mirroring
less or more optimistic future expectations are calculated. The base case scenario is
the most likely scenario. Simulation analysis allows the quantification of probability
distributions of the output as it “uses a random selection of scenarios (likely as well
as unlikely) to generate information”.93
Sensitivity and scenario analyses are often used in real-life applications. The sensi-
tivity analysis is seen as the most popular method in applied work.94 However, both
methods have significant drawbacks. The sensitivity analysis bears the problem that
it focuses on only one input factor; it ignores interactions among the input factors
and combined effects. The scenario analysis usually ignores correlations between the
single input parameters as well.95 Furthermore, not every possible future state is
taken into consideration and no probabilities for the single scenarios are available;
in particular the worst and best case scenario do not provide information on the
probability of outcomes in these (extreme) ranges.
Figure 2.6 illustrates the drawback of valuation results obtained through the sce-
nario analysis method. These results are compared with results obtained through
simulation analysis. The y-axis of the graph displays the probability, the x-axis the
estimated cash flow. The three scenarios normally provided in a risk analysis are
depicted in figure 2.6 as well. In addition, two probability distributions of the cash
flow, obtained by simulation analysis, are contained in the figure.
When comparing the results obtained by the scenario analysis with the two proba-
bility distributions it stands out that in the case of the first distribution the results
seem to be quite reasonable. The base case is around the mean value of the dis-
tribution and the worst and best case are located near the left and right boundary.
However, in the case of the second distribution both extreme scenarios correspond to
cash flow outcomes that are in the extreme tails of the distribution, being not very
probable. Thus, when the results of the scenario analysis are used in the case when
the real distribution of the cash flow equals the second distribution it is to assume
that the risks, and the potential profit chances, are by far overstated.
Despite this apparent advantage of simulation analysis this method is used sparsely in
real-life applications. Except for financial institutions using highly advanced mod-
93
van Groenendaal & Kleijnen (2002), p. 9.
94
Cf. van Groenendaal (1998), p. 202.
95
Cf. Balcombe & Smith (1999), pp. 118-119.
30
Figure 2.6
Scenario Analysis versus Simulation Analysis
Source: Own work.
els based on simulation analysis to value derivatives contracts and other complex
financial products, the proliferation of simulation analysis is low. This is probably
due to the fact that this method is seen as too complex and not too easy to im-
plement in standard software packages.96 Also it can be presumed that an average
manager is reserved towards stochastic modeling and has a lack of understanding
this method.97 However, the advantages of the simulation analysis compared with
the other two methods are significant. Through the generation of probability distri-
butions this method uses all available information and provides the most meaningful
valuation result.
2.1.2.5 Stochastic Cash Flow Modeling
Stochastic modeling is based on repeated random sampling, i.e. the repeated drawing
of random variables or numbers98 , with the aim to construct a probability distribu-
tion of the output of interest. Monte Carlo simulation (method) and the above
96
In real-life applications, mostly Microsoft Excel is used for cash flow modeling. In principle it
is possible to develop a stochastic cash flow model within Microsoft Excel. However, it is not really
applicable for meaningful computations due to its low computation power.
97
Cf. Kwak & Ingall (2007), p. 46.
98
Random numbers are numbers that exhibit statistical randomness. Cf. chapter 3 for a discus-
sion of a potential computational procedure to obtain random numbers.
31
stated term simulation analysis are commonly used as synonyms.99 I will use the
term stochastic modeling in this dissertation. Since I intend to focus on the stochas-
tic modeling of cash flows I do not discuss stochastic modeling in general; I focus on
stochastic cash flow modeling.
Stochastic cash flow modeling refers to the forecasting of future cash flows with the
aim of obtaining probability distributions of the cash flows at some future point-in-
time. The main advantage of this method is the calculation of probability distribu-
tions instead of point estimates. In addition, this method uses almost all available
information.
Stochastic cash flow modeling can simplified be seen as consisting of three steps:100
1. the specification of the cash flow equation and the input factors,
2. the simulation of future input factor values and cash flow values, and
3. the aggregation of the results to a probability distribution.
The specification of the input factors is based on the underlying cash flow equation
that is used in the second step to calculate the single cash flows. It has to be decided
which input factors are deterministic and which stochastic. Normally, the mean
and the variance of the stochastic input parameters are defined, whereat most input
parameters (due to simplification issues) are characterized by a normal distribution.
In addition, but not necessarily, correlations between the single input parameters can
be specified. A possibility to obtain reliable parameters estimates is to conduct an
expert survey.101 Another possibility is a time series analysis of the historical time
series and a forecast of the relevant parameters. The advantage of the time series
framework is the possibility to directly address the correlation structures within the
input parameters.102 I will discuss the input parameter specification in chapter 3
and the impact of correlations on the modeling results in chapter 5.
The underlying cash flow equation specifies the relationship between the input fac-
tors and the cash flow. The calculation of the cash flow is based on the repeating
random combination of the input parameters. Hacura et al. (2001) state that “dur-
ing the simulation process, random scenarios are built up using input values for the
project’s key uncertain variables, which are selected from appropriate probability
distributions. The results are collected and analyzed statistically so as to arrive at
a probability distribution of the potential outcomes of the project and to estimate
99
Monte Carlo refers to the famous state near France, known for its casinos. In a nice way the
term points out that the component chance is playing an important role within this method.
100
For a discussion of the risk analysis process cf. for example Backhaus et al. (2003).
101
Cf. Hughes (1995), p. 180.
102
Cf. Hui et al. (1993), p. 270.
32
Figure 2.7
A Probability Distribution Obtained by Stochastic Cash Flow Modeling
Source: Own work.
various measures of project risk”.103 The appropriate probability distributions of

the forecasts of the input parameters are based on the provided specifications of
the mean levels, variances, and correlations. The forecast of the future probability
distribution function of all factors that influence the cash flow is a critical task. It
will be discussed in the next chapter.
Finally, the aggregation of the single cash flows to a probability distribution yields a
result similar to the example shown in figure 2.7. The y-axis displays the probability,
the x-axis the cash flow.
The probability distribution in figure 2.7 is similar to a lognormal distribution, a
distribution commonly used in finance.104 However, the elevation on the left side
in the probability distribution depicts one of the advantages of stochastic cash flow
modeling, namely the possibility to model non-analytical probability distributions.
A constellation resulting in the probability distribution shown in figure 2.7 would,
for example, reflect the introduction of political risks. The assumption would be that
a certain (small) probability exists every year that the project must be stopped due
to a change of the political or regulatory environment. The results would lead to an
extreme low, but still positive, cash flow in this year.
103
Hacura et al. (2001), p. 551.
104
Many asset models assume that asset returns follow a geometric Brownian motion. This as-
sumption results in a lognormal distribution of the asset prices. Cf. Hull (2008) for a further
discussion.
33
The advantages and drawbacks of stochastic cash flow modeling are summarized
below:
First, stochastic cash flow modeling overcomes the drawback of point estimates ob-
tained by other risk analysis methods. By random sampling from several probability
distribution functions of the input parameters a probability distribution of the cash
flows at some future points-in-time is obtained; all possible future states are covered.
The obtained probability distributions allow the quantification of the probabilities
of certain future events. Thus stochastic cash flow modeling is, as stated by Kwak &
Ingall (2007), “an extremely powerful tool when trying to understand and quantify
the potential effects of uncertainty of the project”.105 Second, stochastic cash flow
modeling is helpful when comparing capital investments with similar mean NPV but
different risk structures; the obtained probability distributions of the cash flows and
of the NPV allow to take more informed investment decisions.
However, this method also has several drawbacks.106 First, stochastic modeling de-
pends on accurate parameter forecasts and in particular on an accurate specification
of the correlation structure between the input factors. Even small deviations of the
input can have a significant effect on the results. Second, the complexity of the
accurate specification of all input parameters is high, in particular when the correla-
tion structure must be specified. Third, the necessary computing power to complete
the stochastic modeling can be immense.107 Fourth, easy to handle software tools
are still rarely available. In addition, stochastic modeling should only be used when
analytical solutions are not available.108
Hess & Quigley (1963) were the first to apply stochastic modeling in finance.109
Hertz (Hertz (1964a), Hertz (1964b)) suggested that employing simulation tech-
niques instead of single-point estimates of future income and expenses is superior
when appraising capital expenditure proposals under conditions of uncertainty. The
author argues that potential risks are identified more clearly and hence the expected
return is measured more accurately. Smith (1994) outlines how simulations may
assist managers in choosing among different potential investment projects. Spinney
& Watkins (1996) apply stochastic modeling as an approach for integrated resource
105
Kwak & Ingall (2007), p. 49.
106
Cf. Nawrocki (2001) for a discussion of the problems with Monte Carlo simulation.
107
However, this problem is due to advances of the computer technology over the past decades to
become less important.
108
Cf. Nawrocki (2001), p. 93.
109
The first documented application of stochastic modeling was performed for nuclear reaction
studies in the Los Alamos National Laboratory (Metropolis & Ulam (1949), Metropolis (1987)).
Hammersley & Handscomp (1964), followed by Glynn & Witt (1992), were the first to discuss the
relationship between accuracy and computational complexity of Monte Carlo simulations. Since the
Monte Carlo simulation is computationally highly intensive and hence reluctant on fast calculating
machines, its application for the solution of scientific problems significantly rose as the costs for
powerful processor decreased dramatically over the last two decades.
34
planning at electric utilities. They find that this approach offers advantages over
more commonly used methods for analyzing the relative merits and risks embod-
ied in typical electrical power resource decisions, particularly those involving large
capital commitments. More recently, Kwak & Ingall (2007) discuss stochastic mod-
eling for project management and conclude that it is a powerful tool to incorporate
uncertainty and risk in project plans.
2.1.3 Project Finance Valuation
2.1.3.1 Project Finance Specifics
The strong dependence of the debt and equity providers on the project’s cash flow is
the key element of project finance elaborated in the sections above and elementary
for the motivation of this dissertation. Though this is also true for equity providers
when investing in differently financed projects, this situation is not common to debt
providers. The partial takeover of business risk by the debt providers in project
finance forces them to apply valuation techniques different to those in the traditional
loan process. The main focus is on the forecast of the expected future cash flows. An
appropriate forecast becomes critical for the realization and the success of a project
finance investment.110 The fact that the physical assets and future cash flows of
project are of little value in the case of a project failure intensifies the situation.
The traditional loan process focuses on the lending company and not the project for
which financing the loan is required. For the estimation of the company’s credit-
worthiness the debt providers normally collect fundamental data (capital structure,
profitability, etc.) on the company and calculate financial ratios, e.g. the financial
leverage or the cash-to-debt ratio. Based on these historical measures a prognosis
of the future financial situation of the company is made and the risks regarding
the loan repayment are evaluated. The default probabilities estimated within this
process concern the company and not the financed project.111
The characteristics of project finance, namely the non-recourse character of the debt,
the foundation of the SPV with the sole aim of conducting the capital investment and
the usually high specificity of the project, result in a situation for which historical
data are not available. Thus, an evaluation of the capital investment based on the
future cash flows is the only possibility. The creditworthiness of the sponsors is not
of interest but rather the prospect of success of the capital investment.
An important advantage of project finance considering the valuation process is the
110
Cf. Böttcher & Blattner (2010), p. 3.
111
A standard approach for the estimation of the default risk is the use of a commercial credit
rating and the comparison with the historical default rates in the corresponding rating class.
35
creation of the SPV and the limited economic life of the project. This allows an
evaluation of the project on a stand-alone basis.112
Most cash flow models used in reality are based on a fundamental analysis. How-
ever, the capabilities of stochastic cash flow modeling are significant, both from a
theoretical and from a practical perspective. Thus, although stochastic modeling
is rather more complex, it is the appropriate method for a detailed analysis of a
capital investment; other techniques use simplifications that may substantially bias
the outcome. The main advantage of stochastic modeling compared to other risk
analysis methods is its ability to quantify the ex ante default probabilities.113
I aim to apply stochastic cash flow modeling in this dissertation in the context of
project finance to quantify both the default probability (relevant for equity and
debt providers) and the expected profitability of the capital investment (relevant
for equity providers). Furthermore, cover ratios which are of high interest to the
debt providers, can and will be calculated based on the probability distributions of
the future cash flows. In the following I discuss the (capital) point of views and its
specifics individually.
In reality, it is intended that – in the optimal case – the future (forecast) cash
flows are sufficient for the timely repayment of the interest and principal and an
appropriate – risk-adjusted – return for the sponsors.
2.1.3.2 Debt Perspective
Not only the approval but also the pricing of a project finance loan depends on the
default probability and the level of variability of the future cash flows.114 Thus, the
quantification of the default risk and of the cash flow variability is of highest priority
for the debt providers. A minimization of these measures through risk management is
targeted.115 Furthermore, it is of high importance for the debt providers to determine
whether the expected cash flows are sufficient for the interest and principal payments
in every period.116 The methods applied for this purpose can be grouped in static
and dynamic methods.
Cover ratios are the static methods mainly applied by debt providers in the case
of project finance investments. In general, a cover ratio is defined as the quotient
of (cash flow) inflow measure and a measure related to the debt capital. For the
112
Cf. Borgonovo et al. (2010), p. 227.
113
Cf. Werthschulte (2004), p. 2.
114
Cf. Corielli et al. (2008), p. 4.
115
Cf. section 2.1.1.5 for a discussion of the importance of risk-sharing within a project finance
transaction.
116
Cf. Werthschulte (2004), pp. 43-44.
36
calculation of a cover ratio the application of a fundamental cash flow model is

sufficient, i.e. no stochastic cash flow modeling is required. However, the calculation
of cover ratios based on probability distributions of one of the input measures is
possible as well. In this case the mean value of the probability distribution is used
as a point estimate for the cover ratio calculation. The whole information contained
in the probability distribution can also be used as discussed below.
The following three cover ratios are frequently used for project finance investments:
1. Debt service cover ratio (DSCR),
2. Loan life cover ratio (LLCR), and
3. Project life cover ratio (PLCR).
The definitions of these ratios are not consistent in the literature. I hence provide a
general definition.
I define the DSCR as:
CFt
DSCRt = (2.3)
DRAt
with DSCRt the DSCR in period t, CFt the (operative) cash flow in period t and
DRAt the debt repayment amount (DRA), i.e. interest + principal, in period t.
The LLCR and PLCR are defined as:
N P V all cash f lows over lif etime loan

LLCR = (2.4)
Loan Amount
N P V all cash f lows over lif etime project

P LCR = (2.5)
Loan Amount
When these ratios are calculated within a risk analysis the debt providers mostly
demand that the ratios never decline below certain critical values during the whole
lifetime of the loan (project). The critical values depend on the risk, industry, and
other specifics of the project. After the granting of the loan, minimum values of the
cover ratios are often specified as covenants.
For the DSCR, for example, a continuous value above one is critical to guarantee the
ability of the SPV to pay the interest and principal on time. A high DSCR value
also indicates a low default probability. Thus, it can be assumed that higher DCSRs
imply lower interest rates for the granted debt capital. The DSCR is seen as the
central ratio in a project finance investment and usually a DSCR of at least 1.3 is
expected.117
117
The assumed minimal value of 1.3 for the DSCR is based on discussions with project finance
37
The DSCR is a period specific ratio. The other two ratios, namely the LLCR and
the PLCR118 , affect the whole lifetime of the loan (project). Both ratios aim to
determine the project’s ability to repay the loan, only the timeframes are different.
A main drawback of these ratios is that the chronological order of the cash flows is
ignored. The PLCR does not even indicate whether the sum of the cash flows over
the lifetime of the loan is sufficient for the whole repayment.
Further cover ratios are used in project finance investments. I intend to implement
four additional cover ratios in the project finance valuation tool that is developed
within this dissertation.
First, two interest coverage ratios are implemented:
EBITt
EBIT interest coveraget = (2.6)
Interestt
EBIT DAt
EBIT DA interest coveraget = (2.7)
Interestt
with EBITt the earning before interest and taxes (EBIT) in period t and EBIT DAt
the earning before interest, taxes, depreciation and amortization (EBITDA) in period
t. Interestt is the interest expense in period t.
Second, I implement two total debt ratios. One with the funds from operations
(FFO) in the numerator
F unds f rom operations

F F O/T otal Debt = (2.8)
T otal Debt
and one with the free operating cash flow (FOCF) in the numerator
F ree operating cash f low

F OCF/T otal Debt = . (2.9)
T otal Debt
The reader is referred to standard literature on financial statement analysis for fur-
ther details on these ratios.119
The main drawback of cover ratios which are based on point estimates of the (cash
flow) inflow measure is the limited information value of the results. A high DSCR,
for example, only indicates that a default is unlikely. It is not possible to quantify
the remaining default probability. Stochastic cash flow modeling can overcome this
drawback though the providing of probability distributions of the cover ratios.
investors.
118
The PLCR is also known as the Net Present Value Cover Ratio (NPVCR) in the literature; cf.
for example Werthschulte (2004), p. 45.
119
Cf. for example Penman (2007).
38
Dynamic methods applied for the valuation of project finance investments include
the three risk analyses methods discussed above. Since I have already discussed
these methods I focus in the following on the advantages of stochastic modeling for
project finance as it is applied in the context of the later introduced project finance
valuation tool.
The advantages of stochastic cash flow modeling in project finance valuation are:
First, the obtained probability distributions can be used to evaluate the effects on
the cover ratios, i.e. a probability distribution of the cover ratios can also be derived.
The above discussed drawback of cover ratios based on fundamental analysis is cor-
rected. Second, after calculating of the probability distributions of the cash flows in
some future points-in-time120 , the debt providers will be primarily interested in the
quantification of the downside risks. A method to quantify this risk is the cash-flow-
at-risk approach121 , a measure similar to the widely used value-at-risk.122 Through
the quantification of certain confidence levels (mostly 1%, 5%, and 10%) this method
enables the quantification of the maximum shortfall. Third, the probability distribu-
tions of the future cash flows obtained through stochastic cash flow modeling allow
the quantification of the probabilities of cash flows becoming negative in a certain
period. Based on these measures the ex-ante default probability can be estimated.
This is one of the main advantages of stochastic cash flow modeling.
Overall the main advantage of the use of stochastic cash flow modeling from a debt
provider’s point of view is the fact that it allows the exact quantification of certain
risks. However, the user of this approach should always be aware that the results
are based on assumptions. So even if the results seem to be exact, they need to be
interpreted with caution. Two main aspects must be kept in mind. First, stochastic
modeling assumes – similar to the most applied modeling methods – that future
developments can be derived from past values. Second, the results of stochastic
modeling are very sensitive to the input parameters. A misspecification of only one
of the normally dozens of input parameters has the potential to significantly bias the
results.
2.1.3.3 Equity Perspective
As stated by Graham & Harvey (2001) the NPV and IRR method are the two valu-
ation approaches for equity investments most often used in real-life applications.123
120
The future points-in-time ideally correspond to the dates when interest and principal payments
by the SPV are due.
121
Cf. Andren et al. (2005) for an overview on cash-flow-at-risk approaches.
122
However, the cash-flow-at-risk is a top-down approach as the value-at-risk is, in general, a
bottom-up approach; cf. Chiu (2007), p. 2.
123
Cf. Graham & Harvey (2001), p. 196.
39
Discussions with practitioners active in project finance confirm that these two meth-
ods are also mainly used by sponsors in project finance investments. Thus, I assume
that the maximization of the NPV or rather the IRR of a project is in the scope of
the sponsor’s analysis in project finance. In reality, the decision to invest is often
dependent on the reaching of certain thresholds in the measures.
As the sponsors are mainly interested in the estimation of the profitability of a
project they will generally use – similar to the debt providers – a cash flow model to
estimate the future cash flows. The expected NPV and IRR will then be calculated
based on the cash flow projections. Discussions with practitioners active in project
finance confirm the assumption that mainly fundamental cash flow models are used
in reality. Sensitivity and scenario analysis are the risk methods which are generally
applied. However, as it was already discussed above, this dissertation focuses on the
application of stochastic cash flow modeling. I believe that the advantages of this
method significantly exceed its disadvantages.
The probability distributions of the NPV and of the IRR are calculated based on
the estimated future cash flow distributions. In addition, the cash-flow-at-risk is of
interest. The calculation of the cash-flow-of-risks allows sponsors to quantify the
potential shortage of cash flows in certain periods (at a certain level of confidence).
The sponsors can use this measure to estimate the probabilities and amounts of
potential reserve liabilities.
2.1.3.4 Related Literature
Considering its importance in practice, the academic literature on project finance is

surprisingly sparse with only a few relevant papers being published in recent years.124
Most of these papers focus on qualitative, organizational, or legal aspects of project
finance. Regarding the valuation of project finance the number of publications is
even sparser. To my best knowledge only two published papers by Esty (1999) and
Gatti et al. (2007) and one working paper by Dailami et al. (1999) highlight the issue
of (quantitative) cash flow modeling in the context of project finance.
Esty (1999) discusses methods for improving the valuation of project finance invest-
ments from an equity provider’s point of view. To achieve this goal, the author
suggests both advanced discounting methods and valuation techniques. The author
discusses the arising problems from the fact that the leverage of a project finance
investment changes over time, resulting in the fact that the use of a single discount
rate over the whole lifetime of the project is inappropriate for the valuation. Fur-
thermore, the problems with the accurate measurement of leverage, i.e. the question
124
Cf. Esty & Megginson (2001) for an overview on the academic literature on project finance.
40
whether book or market values of debt are appropriate when estimating the value of
debt, are addressed. In addition, the author discusses the use of Monte Carlo sim-
ulation to analyze the uncertainty of cash flows and the advantages of real options
analysis. However, the focus of the work is on the improvement of the discount rate
estimation and not the cash flow forecasts. Thus, besides discussing important ele-
ments for valuation, the author does not present a self-contained model that enables
equity providers to calculate either the profitability of a project finance investments
or the expected probability of a project default.
Dailami et al. (1999) introduce in their working paper a computer-based risk man-
agement tool with the focus on infrastructure project finance transactions. The tool
is capable of analyzing the impact of certain risk factors on a project. The authors
aim that the tool raises awareness and expertise in the application of techniques
related to risk management. The developed tool is able to generate probability dis-
tributions of the project’s NPV, IRR, and other key decision variables. To illustrate
the capabilities of the tool the authors apply it to a coal-fired facility.
Gatti et al. (2007) aim to estimate the value-at-risk of project finance investments.
This risk measure is in particular of interest for banks that are involved in the project
finance business, rating agencies, and regulators. A value-at-risk measure is intended
to support the process of credit risk estimation in a project finance investment. The
authors suggest a model that is based on Monte Carlo simulation to measure the
value-at-risk. The authors describe the process to model the cash flows of a project
in four steps, namely (i) the definition of a suitable risk assessment model, (ii) the
definition of the project variables and key drivers, (iii) the estimation of the input
variables and the respective value distributions and the accounting for correlations
among the variables, and (iv) the modeling of the project cash flows, the calculation
of the outputs, and the valuation results. In addition, Gatti et al. (2007) discuss
the definition of default risk in project finance and how a loss distribution could be
derived. The focus of the authors is mainly on the implication for the lenders.
41
2.2 The German Electricity Wholesale Market
The liberalization of the German electricity market started around 15 years ago.
One of the first visible outcomes was the foundation of two electricity exchanges
which merged to the EEX in 2002. The EEX is, like every other electricity ex-
change, a wholesale market. Thus, it is characterized by a limited number of market
participants and relatively low liquidity levels. These characteristics pose questions
regarding market efficiency and the reliability of the established prices.
In this section I examine both the German electricity market and the German elec-
tricity exchange. I discuss the liberalization of electricity markets and address the
electricity market reform in the European Union and the liberalization process of the
German market. The establishment of electricity exchanges as a result of the liberal-
ization is presented. I subsequently summarize the stylized facts on the commodity
electricity and on electricity prices. In particular, the unique characteristics of elec-
tricity prices are crucial for the empirical analysis of this dissertation. I then focus on
the German electricity market and examine the market size, the market design, and
the market structure. I distinguish between the five market segments which are or
were in existence, namely the intraday market, the block contract market, the day-
ahead market for the spot market, and the futures market and the options market
for the derivatives market. In addition, the evolution of the retail electricity price is
discussed. At the end of the section, I address the EEX and its specifics. I devote
considerable attention to the individual market segments, the traded products, and
the trading process. Moreover, the market participants are examined and the evolu-
tion of the wholesale electricity price over the last years is discussed. Furthermore,
potential drivers of the observed evolution are illustrated.
• What are the unique characteristics of the commodity electricity and of elec-
tricity prices?
• What is the current structure and size of the German electricity market?
• What are the fundamentals of trading on the German electricity exchange and
why are the established prices so important?
42
2.2.1 Liberalization of Electricity Markets
2.2.1.1 Motivation and Market Reform
The liberalization of electricity markets in Europe started in Norway and in Great

Britain at the beginning of the 1990s. The beginning of the liberalization process
was based on the insight that markets are a better allocation mechanism than the
then existing system, a highly regulated market with monopolistic structures on the
supply side.125 The aim was the introduction of free markets and the transformation
of the cost-based regulation into a market-oriented price formation.
Prior to the liberalization, electricity markets all over the world were character-
ized by regulation and by lack of competition. The supply side consisted mainly of
state-owned, vertically-integrated companies or regulated private companies, often
monopolists within their supply area.126 Based on the lack of public markets and
competition the pricing mechanism was intransparent and often derived from a cost-
based approach. The results were missing incentives for cost minimization in both
the electricity generation and the distribution process. Higher costs were – after
an approval of the regulatory bodies – simply passed on to the customers.127 The
regulatory bodies mainly aimed at ensuring stable electricity supply; minimal costs
were not in the focus.
The stated rationale for this market structure was, together with the natural monopoly
character of electricity transmission, the observable high economies of scale in the
electricity industry. Over the years this established the opinion that this market
structure is legitimate and the best available option.128 In addition, the security
of supply is seen as crucial for the development of an economy; hence governments
generally try to keep a certain level of control over the national electricity mar-
ket.129 However, supporters of liberalization tried to unbundle this industry by
asking whether there are at least parts within the value chain where competition is
possible.130
The third election victory of the Thatcher government in 1987 put the privatization
of the British energy industry onto the political agenda, resulting in the passage of
the Electricity Act in 1989; a first and fundamental step into a competitive market.
Shortly after, the Norwegian Parliament started a reform of the electricity market
which was implemented in 1991.131
125
Cf. Weigt (2009) for a recent review of the worldwide liberalization of electricity markets.
126
Cf. Weigt (2009), p. 3.
127
Cf. Kramer (2002), p. 1.
128
Cf. Weigt (2009), p. 3.
129
Cf. Müller-Merbach (2009), p. 6
130
Cf. Wawer (2007), p. 5
131
Cf. Weigt (2009), pp. 4-6.
43
Based on these markets’ experience with liberalization, the European Commission

began reconsidering its electricity policy. The liberalized market gradually became
the new paradigm. The actual liberalization process in Europe began with the
Electricity Directive 96/92/EC in 1996. The directive required the national govern-
ments to develop plans for the liberalization of their respective markets by February
1999.132 The directive included the introduction of a freedom of choice of the electric-
ity supplier for a certain share of customers. In addition, it proposed three potential
third-party access models to the transmission networks. It also required the admin-
istrative unbundling of supply, generation, and network activities.133 The intention
was to gradually open the markets and in particular to provide third-party access to
the networks, an obligatory condition for competition.134
In 2001, the European Council concluded further measures were necessary, as the
outcome of the first directive had been widely regarded as unsatisfactory. The Euro-
pean Council thus approved a second directive, the Electricity Directive 2003/54/EC.
This directive reduced the freedom of choice for the national governments and short-
ened the deadlines. According to the second directive, the market for non-household
electricity consumers and for all consumers had to be liberalized by certain point-in-
times.135
2.2.1.2 Liberalization of the German Market
Prior to the liberalization of the German market, the regulation had been based on
the 1936 federal Energy Industry Act, the Energiewirtschaftsgesetz (EnWG).136 The
preamble of the EnWG disclosed the law’s intention by stating that the damaging
economic drawbacks of competition must be avoided in the electricity industry in
order to secure cheap electricity supply. The results of the regulation were closed
supply areas and an electricity market being a private sector under state supervision.
In the German electricity market the starting point of the liberalization process was
the first European Electricity Directive of 1996 which was implemented in the federal
Energy Industry Act in 1998. The act’s intention was to open electricity supply for
competition. Main issues were – as requested by the European Commission – free
selection of the electricity supplier for end consumers and the formulation of rules
of third party access to the transmission networks.137 However, the act allowed the
industry a wide spectrum of choices of how to implement the required liberalization
132
Cf. Polo & Scarpa (2002), p. 18.
133
Cf. Weigt (2009), p. 8.
134
Cf. Krisp (2007), p. 38.
135
Cf. Weigt (2009), pp. 8-9.
136
Cf. Krisp (2007) for a discussion of the German electricity policy since the 1980s.
137
Cf. Ockenfels et al. (2008), p. 4.
44
steps.138 This freedom of choice resulted in an insufficient self-regulation of the

market; the intended level of competition was not reached. Thus, after the second
directive, the German government tightened the requirements.
One of the new measures was the establishment of a regulatory body, the Federal Net-
work Agency (Bundesnetzagentur). The Federal Network Agency took up regulatory
activity in the field of third party access and implemented rules for this procedure.
The charged fees for the use of the transmission network by third parties, for exam-
ple, are now subject to approval by the agency. Furthermore, the unbundling of the
vertically integrated supply companies was demanded by law in 2005.139
2.2.1.3 Establishment of Electricity Exchanges
The liberalization process drove the need for marketplaces, resulting in the founda-
tion of electricity exchanges. The foundation of the Nord Pool in Oslo in 1993 marks
the introduction of the exchange trading of electricity in Europe.
Since the first Electricity Directive of the European Commission did not contain any
specifications regarding the design of the electricity wholesale markets most trading
took place in over-the-counter (OTC) markets at the beginning. However, due to
the intransparency and slowness of these markets electricity exchanges were soon
founded, in order to provide publicly available reference prices.140
In particular at the end of the last century, a multitude of national electricity ex-
changes were founded all over Europe. Figure 2.8 contains an overview of the elec-
tricity exchange landscape in Europe in 2007.
At the moment almost twenty electricity exchanges exist within the European Union.
Figure 2.8 reveals that besides the Scandinavian countries almost every country has
its own electricity exchange. The common electricity market of the Nordic countries
– Sweden, Norway, Finland, and Denmark – located at the Nord Pool, is a first
example for an international electricity market.141 First steps in the direction of
a market coupling on an international level are observed. Examples are Belgium,
France, and the Netherlands where a (day ahead) market coupling was recently
introduced.142
138
Cf. Weigt (2009), p. 11.
139
Cf. Weigt (2009), p. 11.
140
Cf. Ockenfels et al. (2008), pp. 11-12.
141
The integration of the national Nordic markets to a joint Nordic market had taken place between
1993 and 2000. In 1993 the Nord Pool started operation for the electricity market of Norway; in
1996 it was joined by Sweden and in 1998 by Finland; in 1999 and 2000 the western and eastern
part of Denmark, respectively, completed the joint market.
142
Cf. Meeus et al. (2006) for a discussion of the market integration in France, Belgium, and the
Netherlands.
45
Figure 2.8
Electricity Exchange Landscape in Europe in 2007
Source: Taken from Menzel (2007), p. 10.
On most electricity exchanges trading takes place in spot and futures markets. Op-
tion markets also exist but the liquidity in these markets is normally very low.
Contracts with physical and financial settlement are traded on almost all exchanges.
The main part of trading is observed in the futures markets, where the main part
of the liquidity is observed in futures with financial settlement.143 All electricity
exchanges serve as wholesale markets.
Initially the prices on the single electricity exchanges were largely independent and
unaffected by events in neighboring electricity markets. However, a change of the
situation could be observed over the last years. Empirical research suggests that
the correlation between the markets is growing, even though the pace is differently
evaluated.144 In particular the market coupling efforts undertaken by the European
Commission seem to strengthen the market integration.
The importance of the electricity exchanges and of the established prices is enormous
as they establish a reference price. Assuming a sufficient level of liquidity in the
exchange-based trading, no electricity buyer or seller will agree to a bilateral contract
143
Cf. section 4.4.1 for a discussion of the liquidity in the single market segments of the German
electricity exchange.
144
Cf. for example Armtrong & Galli (2005), Zachmann & von Hirschhausen (2008), Bundesnet-
zagentur (2010), and Bosco et al. (2010).
46
(disregarding the transaction costs) with a specified price below the actual exchange
price. The electricity prices established on the exchange thus serve as reference prices
for the whole electricity market.145
Due to the high number of national exchanges, many of them characterized by low
liquidity, it is to assume that a consolidation of the electricity exchange landscape
will take place in the next years. The cooperation of the EEX and the Powernext
established in 2008 is one of the first examples for this process.146
2.2.2 Stylized Facts on Electricity and on Electricity Prices
The commodity electricity and the observed electricity prices exhibit unique charac-
teristics which I discuss in the following; I distinguish between stylized facts on the
commodity electricity and stylized facts on electricity prices.
2.2.2.1 Stylized Facts on Electricity
The following stylized facts on the commodity electricity are commonly found in the
existing literature:
1. Non-storability,
2. Grid-bound transportation,
3. Regionalism, and
4. Independence of spot and forward markets.
Non-storability
Non-storability is the outstanding characteristic of the commodity electricity. From a
technical point of view some storage technologies can be identified, e.g. pump water
hydro plants or storage lakes. Pump water hydro plants are the only technology
available to store electricity on a larger scale.147 However, from an economic point
of view it is assumed that electricity is not storable due to the high costs involved.
Most other unique characteristics of electricity are due to its non-storability. It is
also the reason behind the necessity to generate and consume electricity simultane-
ously.
145
146
Cf. 2.2.4.1 for a discussion of this cooperation.
147
Cf. Schill & Kemfert (2009), p. 3.
47
Grid-bound transportation
The transmission of electricity takes place in power grids. They are the only transport
possibility for electricity on a larger scale. The grids often span over thousands of
kilometers and require investments in the magnitude of billions of Euro. In particular
the transmission over long distances requires special networks that are both costly
and sensitive to external influences. The operation of the grids is difficult due to the
fact that input and output always have to be balanced. An in-balance can lead to a
collapse of the whole grid.
The grid-bound transportation is also the cause of another significant problem in
electricity markets: transmission constrains or congestion. Congestion arises when
a grid is operated at its physical maximum. No additional electricity can be then
transported through this grid. The setting of rules about how to handle congestion
is very important for electricity markets.148
Regionalism
Due to capacity restrictions the transmission of large electricity amounts over na-
tional borders is often difficult. Therefore, the available transmission capacity through
the cross-borders cables is auctioned. This takes place in the form of options, the
so called physical transmission rights (PTRs). A trader has to buy a PTR in order
to be able to exploit price differences between two national spot markets. The PTR
must normally be bought before the electricity prices are known; this implies taking
over additional risk. The German power grid, for example, is connected through
three high-voltage cross-border cables with the Dutch power grid.149 Despite the
auctioning of PTRs price differences between the markets seem to prevail.150
The lack of sufficient cross-border capacity is one of the reasons why despite the
worldwide liberalization of national electricity markets an integrated single (Euro-
pean) electricity market still is not reached. The national markets are not coupled,
consisting mainly of regional markets. One of the implications is that price changes
in national market do not impact – at least in the short-term where arbitrage is not
possible – the price in another market or region or vice versa.
148
Cf. for example Furio & Lucia (2009) regarding the Spanish market.
149
Cf. Marckhoff (2009) for a discussion of the German-Dutch cross border market and the trading
of PTRs.
150
Cf. Andeweg et al. (2009) for a discussion of profitable trading strategies based on this obser-
vation.
48
Independence of spot and forward markets

The independence of the spot and forward markets is due to both the non-storability
of electricity and the resulting non-existence of an arbitrage-relation. This charac-
teristic is almost unique as the most asset and commodity spot and forward markets
are linked by an arbitrage relation.151
The result of this characteristic is a loose relation between the spot and forward price,
leading to an extensive research on price formation in electricity futures markets as
it is discussed in the next section and in the empirical part of this dissertation. The
loose relation, among other things, results in a low correlation between the spot and
futures time series; this is even amplified through the high volatility and kurtosis of
the spot prices as it is discussed below.
2.2.2.2 Stylized Facts on Electricity Prices
The following stylized facts on electricity prices are found:
1. Seasonality,
2. High Volatility,
3. Price spikes,
4. Mean reversion, and
5. Negative prices.
Seasonality
Seasonality on a daily, weekly, and yearly scale is observed in electricity price time
series.152 Climatic changes over the year, i.e. temperature fluctuations and varying
numbers of daylight hours, lead to a seasonality on the demand side.153 The demand
is mostly higher in the winter. However, in certain regions a peak in the summer
months due to increased use of air condition can be observed as well. Countries
heavily relying on water power, e.g. the Scandinavian countries154 , also face season-
ality on the supply side due to the fluctuating level of water reservoirs. Both effects
151
Examples for other non-storable commodities are difficult to find as almost every commodity
is storable to some extent. Cattle, eggs and pork bellies are examples for commodities with limited
storability.
152
Regarding the German market, the daily and weekly seasonality is significant. Regarding the
yearly seasonality empirical evidence is mixed. As, for example, discussed by Weron (2006) there is
no clear evidence for a yearly seasonality in the German market.
153
Cf. Weron et al. (2004), p. 2.
154
Around 50% of the electricity generated in the Nord Pool is produced by hydro plants; cf. Bask
& Widerberg (2009), p. 279.
49
lead to a yearly seasonality. The weekly seasonality is explained by the high share
of industrial electricity consumers155 that do not demand electricity on weekends
and non-working days, i.e. by changing business activites. The daily seasonality is
related to the different demand patterns during night and day, again mainly caused
by the industrial consumers.
Not only the mean but also the volatility of electricity prices seems to be time-
dependent.156
High Volatility
Electricity prices are characterized by volatilities merely observed in other markets.
Daily volatilities over 40%157 are observed; volatilities of two orders of magnitude
higher than observed for other asset price time series are common.158 The non-
storability of electricity leads to a situation where variations in supply and demand
almost have real-time influence on the electricity price. In particular changes on the
supply side translate into high increases of the electricity price.
Furthermore an “inverse leverage effect” is observed, indicating that the volatility
of electricity prices tends to rise more strongly after price increases than after price
decreases.159
Price spikes
Price spikes are narrowly linked to the high volatility. Price spikes are a sudden
increase of the electricity price, often more than tenfold. These high prices tend
to remain for hours, at the highest to days, and return afterwards to normal price
levels.160
As no inventories of electricity can be used to dampen shocks in the supply or demand
the results of these shocks are severe.161 Another important factor favoring price
spikes is the inelasticity of the demand. The fatalist price spikes are hence observed
after power plant outrages, when a significant part of the supply side disappears
almost immediately.
155
The distribution over the three largest electricity consumer groups in the German market is as
follows: industry consumers (47%), households (26%), and commercial consumers (14%). Together
with commercial and other consumers almost three fourth of the total electricity consumption in
Germany is due to non-household consumers; cf. Bundesministerium für Wirtschaft und Technologie
(2010) for the data.
156
Cf. Weron (2008), p. 1100.
157
Cf. Ullrich (2009) for a discussion of the importance of data frequency when analyzing time
series characteristics of electricity prices.
158
Cf. Weron (2006), p. 101.
159
Cf. Knittel & Roberts (2005), p. 793.
160
Cf. for example Seifert & Uhrig-Homburg (2007) for a discussion of price spikes.
161
Cf. Marckhoff (2009), p. 1.
50
One result of the price spikes is a non-Gaussian distribution of electricity prices. A

significant heavy tail is observed and the resulting distributions are right skewed.
Regarding the electricity price modeling and forecasting price spikes play an out-
standing role on the short-term time scale. On a long-term scale, on the other side,
they are due to the short lifetimes of minor interest.
Mean reversion
Mean reversion refers to the characteristic of electricity prices to revert in the mid-
and long-term to a long-term average price.162
The rationale behind the mean-reversion of electricity prices is the (long-term) ad-
justment of the supply to the demand, resulting in a convergence of the electricity
price to the cost of production.163 Furthermore the mean reversion of the weather as
a dominant driver of the equilibrium electricity price could lead to a mean reversion
of the prices as well.164
Negative prices
Another characteristic of electricity price time series is the occurrence of negative
prices. This is due to the costly and sometimes technically not even possible shut-
down of power plants in the short-term. A power plant shutdown is associated with
high costs. In certain hours it can hence be reasonable for a power plant operator to
pay positive amounts to market participants who are willing to take the produced
electricity from the network.
In such a case the buyer could be another electricity supplier who has the possibil-
ity to shut down his power plant at lower costs or in a shorter time period. The
received electricity is then used to satisfy the existent delivery obligations instead
of generating it oneself. The (simplified) gain of the trade is the received amount
minus shutdown costs.
2.2.3 German Electricity Market
In this section, I address the German electricity market. First, I examine the market
size of both the international and the national market.165 Second, I discuss the
162
Cf. section 3.5.2 for a discussion of this time series characteristic.
163
Cf. Cartea & Figueroa (2005), p. 314.
164
Cf. Escribano et al. (2002), p. 4.
165
All statistical data reported in this section are based on total electricity generation in the cor-
responding year. The data are obtained by the German Federal Ministry of Economics and Tech-
nology, the Bundesministerium für Wirtschaft und Technologie (BMWi); cf. Bundesministerium für
Wirtschaft und Technologie (2009).
51
market structure of the German electricity market. Third, I address the structure of
the supply and the demand side. Fourth, I illustrate the evolution of the electricity
price over the last decade.
2.2.3.1 Market Size
Total world generation of electricity amounted to 19,855 TWh in 2007. The highest
ratio of the generation was contributed by Asia, followed by North America and
Europe. Regarding the European market, Germany (637 TWh) was the largest
electricity generator, followed by France (570 TWh), and by Great Britain (396
TWh).
The German electricity market is both on the generation and on the consumption
side the largest market in Europe. With its total generation of 637 TWh the German
market is around ten percent larger than the French market and is equal to around
12% of the total electricity generation in Europe. However, when compared with
the world market the German market accounts for only 3.2% of total worldwide
generation. Compared with the two largest electricity producers in the world, the
United States (22%) and China (17%), the size of the German market is relatively
moderate.
Due to an under-proportional growth of the German market in relation to the total
market the German market share on the worldwide market is supposed to further
decline in the future. Over the time period 1990 to 2007, for example, the com-
pounded annual growth rate of the world market was around 3.1%. The German
market grew with an average growth rate of approximately 0.9% over this period.
2.2.3.2 Market Design
Two principal market structures are found in liberalized electricity markets. On the
one side, the pool model as a centralized model, on the other side, the exchange
model as a decentralized organization form of the market. The pool model is mostly
implemented in English-speaking countries; the exchange model is deployed in all
European countries except the British market.166 In addition, bilateral trading (OTC
markets) – mainly for forward transactions – is found in most national markets.
The German electricity market is organized based on the exchange model. It consists
of a sequence of markets on which electricity for delivery in a certain time period is
traded at different times. The main market is the wholesale market with its market
166
Cf. Grimm et al. (2008) and Ockenfels et al. (2008) for a discussion on the two market structures
and how they are implemented in the largest European markets.
52
segments (intraday market, day-ahead market, and futures market). The German
wholesale market is located at the EEX. The EEX is discussed in section 2.2.4.
The exchange model exhibits a central challenge: The co-ordination of the markets
for generation, transmission, and balancing energy.167 Regarding the German mar-
ket, the transmission within the national borders is so far unproblematic since the
capacity of the German power grid is sufficient.168 The balancing energy market169
(sometimes also control energy market) serves the purpose to secure the stability of
the power grid. It is operated by the transmission system operators (TSOs). The
TSOs are responsible for the permanent balance between electricity generation and
demand in their control areas170 ; their assignment is to secure that the power fre-
quency remains stable.171 An unstable frequency mostly implies a breakdown of the
grid. Parts of the balancing energy are auctioned on a daily basis, with the grid
operators on the demand side and the power companies and electricity traders on
the supply side.172
The value chain within the power sector is classified by thee broad levels: (1) gener-
ation, (2) transmission, and (3) distribution. The power plant owners, in Germany
normally one of the four large utilities173 , are responsible for the electricity gener-
ation. The transmission of the electricity over the grids is organized by the TSOs.
The fee for the transmission in the case of an inequality between power generator
and owner of the TSO is approved by the German regulator, the Bundesnetzagen-
tur.174 In the majority of cases the distribution of the electricity is lastly performed
by public services.
Regarding the importance of the wholesale market, it is to assume that similar to
the time before liberalization still a high percentage of the supply contracts in the
German market is characterized by long maturities and fixed prices.175 Furthermore,
it is to assume that a high percentage of the electricity that is traded between the
generation units and the distributors has long-term fixed prices. However, it is also
to assume that the market participants do not hedge all of their price risk. A certain
percentage will be left open and bought short-term at the wholesale market at the
167
168
Cf. section 2.2.2.1 for a discussion of the electricity transmission over national borders.
169
Thee different types of balancing power are distinguished, namely primary, secondary, and
tertiary balancing power. The differences are the activation times and the durations of operation;
cf. Flinkerbusch & Heuterkes (2010) for more details.
170
The German market consists of four control areas. The Bundesnetzagentur recently postulated
to pool these four control areas due to cost reduction potentials; cf. Bundesnetzagentur (2010).
171
Cf. Flinkerbusch & Heuterkes (2010), p. 4713.
172
Cf. for example Rammerstorfer & Wagner (2009) for a discussion of this procedure.
173
Cf. the next section for a discussion of the structure of the supply side.
174
Cf. for example Dieckmann (2008) for a discussion of the transmission process.
175
Regarding the North American market Nakamura et al. (2006) report that about 50% of all
electricity deliveries are still determined by bilateral forward contracts.
53
actual market price. The wholesale market price then serves as a decision foundation
and the prices of new long-term contracts and the wholesale price converge. Thus,
the wholesale market determines the whole market (in the long-term) as all contracts
reflect the wholesale market price.176
2.2.3.3 Market Structure
In 2007 the total German generation capacity amounted to 137.5 GW. Anthracite
(29.3 GW), lignite (22.5 GW), wind (22.2 GW), and nuclear power (21.3 GW) were
the four fuels with the main ratio on the German total generation capacity.
When analyzing the German electricity capacity and generation statistics, the de-
composition of the year’s generated electricity shows that the ratios of one fuel on
total capacity and on total generation can differ significantly. According to the offi-
cial statistics nuclear power was the most important fuel in the German market in
2007. 27% of the total generation was due to nuclear power; lignite and anthracite
followed by 26% and 22%, respectively. The electricity generation in Germany in
2007 was hence characterized by the use of fossil fuels and nuclear technology. The
three fuels accounted for around three-quarts of the total electricity generation.
The available data further reveal that wind and water accounted for around 7% of
the total generation. When compared with a combined share of 23% on the total
generation capacity this number appears surprisingly low. However, the specifics of
the electricity generation based on water and wind need to be taken into account to
explain this wide gap.
In Germany water based electricity generation is almost solely dependent on pump
water power plants. In these power plants electricity is first transformed in potential
energy. Then it is transformed back to electricity. The costs of this technology are
high. Thus, pump water power plants are, from an economic point of view, more a
storage than a generation technology and tn reality mostly used in peak load hours.
The electricity generation based on wind energy on the other side is very sensitive
to the actual wind situation. A too light or too strong wind results in a switch off
of the wind turbine. The result is a low efficiency, often reaching efficiency degrees
of only 20% or even less.
The composition of the German electricity generation facilities has experienced sig-
nificant changes over the last years. In particular the turn of the millennium marks
a turning point in the German energy policy. At this time, the German govern-
ment passed the Renewable Energy Act, the Erneuerbare-Energien-Gesetz (EEG).
The intention of the law was to increase the share of renewable energy in the Ger-
176
54
man generation mix. The EEG defines fixed and guaranteed prices for electricity
generated by generation technologies based on renewable energies. The result is a
significantly improved risk structure for investments in these generation technologies.
Thus the act provides incentives for the construction of renewable energy generation
capacity. The investment process in renewable energies is stimulated. Finally, the
share of renewable energies on the German electricity generation significantly in-
creased.177 The quantitative aim of the act is an increase of the share of renewable
energies on the total electricity generation to at least 30% in 2020 and a further con-
tinuous increase afterwards.178 However, the economic impact of the implemented
system is under controversial discussion.179
The results of this decision can be already seen in the evolution of the generation mix
over the last years. Starting with 1.1% in 1990, the combined share of the renewable
energies increased from 2.1% in 1999 up to 6.8% in 2008.180
Traditionally, a high concentration is observed in the ownership of the German gener-
ation capacity. At the moment the German market is characterized by the existence
of four large utilities who control almost 80% of the total generation capacity. The
four utilities are E.On (34%), RWE (27%), Vattenfall (11%), and EnBw (7%).181
The German regulator, the Federal Network Agency, assumes that RWE and E.On,
with a combined market share of over 60%, form an oligopoly in the German gener-
ation market.182
2.2.3.4 Retail Electricity Price
Regarding the retail electricity price, two different prices need to be distinguished.
The first is the electricity price charged from household customers, the second is
the electricity price charged from industry consumers.183 The reason for the differ-
entiation are the differences in the demanded amount, the demand profile, and the
distribution channels.
Figure 2.9 depicts the evolution of the two price time series between 1991 and 2007.
The straight line is the household price, the dashed line the industry price.
177
Cf. Wawer (2007), pp. 97-101.
178
Cf. Erneuerbare-Energien-Gesetz (2008).
179
Cf. for example Frondel et al. (2010).
180
With the political decision to step off from nuclear technology the future development of the
electricity generation based on renewable energy becomes even more critical for the German elec-
tricity supply
181
The statistic is based on all power plants with a nominal capacity larger than 100 MW.
182
Cf. for example Bunn & Karakatsani (2008).
183
In general, a classification by the annual consumption is undertaken. According to Eurostat
(2009) household consumers are defined as the consumer band between 2,500 and 5,000 kWh and
industry customer as the consumer band between 500 and 2,000 MWh.
55
Figure 2.9
Evolution of German End Consumer Electricity Price 1991-2007
25
20
Price [Euro/MWh]
15
10
0
1991 1993 1995 1997 1999 2001 2003 2005 2007
Source: Own work, based on data from Bundesministerium für

Wirtschaft und Technologie (2010).
The price evolution in figure 2.9 reveals that beginning with the liberalization in
the mid of the 1990s the electricity price started to decline. However, the phase
of declining prices preserved only for a few years and, after reaching a minimum
in 2000, a phase of increasing prices followed. This phase has continued until now;
the introduction of the emissions right trading in 2005 had no visible impact on
this trend. Thus it seems to be legitimate to pose the question whether the hoped
lowering of electricity prices was not reached by the liberalization.
To answer this question a closer look at the underlying price components of the
consumer price has to be taken. To gain an understanding of the price components
the decomposition of the German private end consumer price in 2007 is used. This
year’s consumer price is affected by seven price components. In decreasing impor-
tance these are (the percentage of the cost factor on the private end consumer price
is found in the brackets):
• electricity generation (37%),
• transmission (24%),
• sales tax (16%),

56
• electricity tax (9%),
• license fees (8%),
• EEG184 (5%), and
• KWK185 (1%).
Regarding to this statistic it is rather difficult to answer the question on the impact
of the market liberalization on the prices. This is due to both the relatively low
share of the electricity generation on the total end price and the introduction of new
price components by the EEG in recent years. Thus, the literature on this topic is
very mixed and the final answer to this question is still outstanding.186
A comparison of the German electricity price with the prices in other European
countries reveals that the German prices are above the average price. According to
the available statistics Denmark has the highest household prices (0.299 Euro/kWh)
and Italy the highest industrial prices (0.1435 Euro/kWh). The average household
price in Europe is 0.1659 Euro/kWh, the average industrial price 0.0987 Euro/kWh.
The prices reported for Germany are 0.2528 Euro/kWh, and 0.1132 Euro/kWh, re-
spectively. The data are based on prices for 2009 and include all taxes. However,
it has to be noted that the above discussion showed that the comparison does not
enable to draw conclusions on the total generation costs. The non-generation costs
in Germany are higher than the generation costs; the same could be observed in
other countries.
2.2.4 European Energy Exchange
2.2.4.1 General Remarks
The German marketplace for electricity is the EEX.187 The EEX is an electronic
exchange which was established in 2002 as a result of a merger between the Leipzig
Power Exchange (LPX) and the former European Energy Exchange, previously based
in Frankfurt am Main.188
184
Erneuerbare-Energien-Gesetz; allocation of cash flows to finance the guaranteed prices – which
are significantly higher than the market price – for electricity that is generated based on renewable
energies.
185
Kraft-Wärme-Kopplung; surcharges for combined heat and power plants.
186
Cf. for example Bonneville & Rialhe (2005) for a discussion of the impact of liberalization.
187
Cf. Soennecken & Pilgram (2002) for a discussion of the newly founded EEX in 2002.
188
The LPX was founded in 1999, the EEX in 2000. The focus of the LPX was on the trading of
hour contracts in the day-ahead market; the focus of the EEX was on the trading of block contracts
in a continuous (block) market.
57
The EEX is based in Leipzig and operated by the EEX AG.189 The EEX sees itself
as an European marketplace. It pursues the strategy of an open business model190
to generate more flexibility, market coverage, and liquidity. This is achieved through
targeted spin-offs and co-operations.191 Being a public sector institution the EEX is
subject to the German exchange law. Also being a derivatives trading place the Ger-
man Securities Trading Act, the Wertpapierhandelsgesetz (WpHG), applies for the
derivatives market. Furthermore the EEX is monitored by the German Federal Fi-
nancial Supervisory Authority, the Bundesanstalt für Finanzdienstleistungsaufsicht
(BaFin).
The tradable commodities on the EEX are coal, electricity, emission rights192 , and
gas193 . Emissions rights, electricity, and gas contracts are traded both in a spot and
in a futures market. Coal is only traded in the form of financial futures. Clearing194
of OTC trades in electricity is also offered by the EEX.195
In the following I solely focus on the electricity segment of the EEX.
The EEX is the largest electricity exchange in continental Europe and – after the
Scandinavian Nord Pool – the second largest in Europe. As of the end of 2009, 191
market participants from 19 countries trade in the spot and futures market of the
EEX.196 The total traded volume in the spot market in 2009 was approximately
203 TWh, in the futures and forward market197 around 1025 TWh.198 Compared
with the above reported statistics for the German electricity market these numbers
translate to a factor of two between the traded and the consumed electricity in the
German market. When compared with the yearly (gross) electricity consumption of
around 600 TWh almost 25% of German electricity is traded in the physically settled
spot market of the EEX.
In 2008, the EEX and the French energy exchange Powernext declared an intense
cooperation regarding their electricity trading activities. The results of this cooper-
189
The EEX AG is a public company which has two large shareholders, the Eurex (European
Exchange) Zürich AG with 34.73% and the Landesbank Baden-Württemberg with 22.64%. The
remaining shareholders mainly include utilities, energy trading companies, and banks.
190
The preamble of the EEX exchange rules states that “the EEX and its holding companies see
themselves as a European market place and aim to develop this market place, e.g. by means of
co-operations”; EEX (2010a), p. 4.
191
Cf. EEX (2010b), p. 1.
192
The trading of emission rights was introduced in 2005 when the European Union Emission
Trading System (EU ETS) was launched.
193
The trading of gas contracts was introduced in 2007.
194
Cf. Soennecken et al. (2002) for a discussion of the OTC clearing on the EEX.
195
In 2006 the EEX AG founded the European Commodity Clearing AG (ECC) as a 100% sub-
sidiary. The ECC now serves as the central counterparty in all trades.
196
Cf. section 2.2.4.4 for a discussion of the market participants on the EEX.
197
Only forwards cleared by the EEX are included in this statistic. It is estimated that the OTC
forward market is multiple magnitudes larger than the exchange trading.
198
Cf. section 4.4.1 for a discussion of the liquidity on the EEX.
58
ation are a joint electricity spot market and a joint electricity futures market. The
spot market – operated by a Societas Europaea, the EPEX Spot SE, of which both
parties held 50% – is now located in Paris.199 The futures market – organized in
the EEX Power Derivatives GmbH200 , majorly owned by the EEX – is located in
Leipzig.
2.2.4.2 Market Structure and Traded Products
The electricity segment of the EEX consists of a spot201 and of a derivatives mar-
ket202 . The spot market is comprised of two market segments, an intraday and a
day-ahead market. This is a market structure which can be found in most electricity
exchanges. The day-ahead market has been active since the founding of the EEX.
The intraday market was introduced in September 2006. In addition, contracts for
three specific blocks of hours have been traded in a block contract market until
August 2008. The derivatives market consists of a futures and of an options market.
Figure 2.10 summarizes the market structure of the electricity segment of the EEX.
I discuss the individual market segments, and the traded products in this market
segments, in the following separately. The focus is here on contracts with delivery,
or rather settlement, in the German market area.203
Day-Ahead Market
Contracts with a delivery period of one hour are traded in the day-ahead market.
These hour contracts204 ensure the delivery of electricity for a specified delivery hour.
The contract size205 is 0.1 MW and the settlement of the contracts is physical.206
The price finding mechanism in the day-ahead market is a uniform auction, a com-
mon and accepted mechanism in electricity day-ahead markets.207 In each auction 24
199
Now electricity for physical delivery in four countries – Germany, France, Austria, and Switzer-
land – is traded on the spot market of the EEX.
200
The derivatives market was spun off to the EEX Power Derivatives GmbH in September 2008.
The foundation of the new subsidiary was conducted respectively as of January 1, 2008.
201
The trading in the spot market is based on the XETRA (Exchange Electronic Trading), the
electronic securities trading system of the German stock exchange operator, Deutsche Börse AG.
202
The trading in the derivatives market is based on the electronic system of the derivatives
exchange EUREX, jointly operated and owned by the Deutsche Börse AG and the Swiss exchange
operator SIX Swiss exchange.
203
The other market areas serviced by the EEX are Austria, France, and Switzerland.
204
Cf. EEX (2008b) and EEX (2008a) for further details regarding the contract specifications.
205
Load profile is another term used for the contract size. Contract size refers to the delivery
rate, i.e. the quantity of electricity per hour. Based on the contract size the contract volume is
determined as the product of delivery rate and the number of delivery hours in the delivery period.
206
The delivery of the electricity can take place in one of the following control areas: Amprion
GmbH, Transpower Stromübertragungs Gmbh, 50Hertz Transmission GmbH, EnBW Transport-
netze, and Austrian Power Grid.
207
Cf. Ockenfels et al. (2008) for an academic review of the trading process in the day-ahead
59
Figure 2.10
Market Structure EEX
ELECTRICITY market
SPOT market DERIVATIVES market
INTRADAY market DAY-AHEAD market FUTURES market OPTIONS market
(BLOCK market)
Source: Own work.
independent prices are established for each hour of the delivery day. Until Septem-
ber 2008, auctions were only taking place from Monday to Friday excluding public
holidays. On Fridays and before public holidays more than one auction accordingly
took place. Starting September 9, 2008 the EEX introduced seven-day-trading in
the day-ahead market.
The predefined price range in the day-ahead market is from minus 3,000 up to 3,000
Euro/MWh. The possibility to bid negative prices208 in the day-ahead market was
introduced in September 2008. The minimal price fluctuation is 0.1 Euro/MWh.
In addition to the hour contracts block contracts can be traded in the day-ahead
market. Besides the standardized blocks209 market participants can define arbitrary
combinations of hours. Block bids are a special bid form that ensures that either all
or none of the specific hour contracts are traded, resulting in an ‘all-or-nothing‘ bid.
The EEX calculates a daily and a monthly index for the day-ahead market, the Phys-
ical Electricity Index (Phelix). The daily index is calculated as a simple arithmetic
average of the hourly prices for the base (0 am to 24 am) and peak hours (8 am to 8
market.
208
Cf. section 2.2.2.2 for a discussion of negative prices in electricity markets.
209
The standardized block orders are: block baseload (hour 1 to 24), block peakload (hour 9 to
20), block night (hour 1 to 6), block morning (hour 7 to 10), block high noon (hour 11 to 14), block
afternoon (hour 15 to 18), block evening (hour 19 to 24), block rush hour (hour 17 to 20), block
off-peak I (hour 1 to 8), block off-peak II (hour 21 to 24), and block business (hour 9 to 16).
60
pm), resulting in the daily Phelix Base and Phelix Peak.210 In addition, a monthly
index is calculated as an arithmetic average of the daily index values. The monthly
Phelix Peak is calculated only based on prices between Mondays and Fridays; non-
working days are ignored. The daily Phelix Base can be seen as the reference price
in the German electricity market.
Intraday Market
In the intraday market hour contracts with similar specifications as in the day-ahead
market are traded. The intraday hour contracts have a contract size of 0.1 MW and
the minimal price fluctuation is 0.01 Euro/MWh.
The permitted price range for intraday market contracts is from minus 9,999 up
to 9,999 Euro/MWh. Negative prices in the intraday market were introduced in
December 2007. Similar to the day-ahead market trading at negative prices occurs
occasionally.211 In contrast to the day-ahead market the intraday market is operated
as a continuous market with trading taking place around the clock.
Block Contract Market

Trading took place in the block contract market until August 2008. Block contracts
ensured the delivery of power over several delivery hours with a delivery rate of 1
MW. Traded block contracts were a base load, a peak load, and a weekend base
load contract.212 The base load block contract ensured the delivery of electricity
throughout the day while the peak load block contract only ensured the delivery in
the peak hours (8 am to 8 pm). The base load block contract was available for all
days, the peak load block contract only between Mondays and Fridays. The weekend
base load block contract ensured delivery in all hours of the weekend.
A price range was not specified in the block contract market. Only a positive price
was required, translating to a minimum price of 0.01 Euro/MWh. The minimum
price fluctuation was 0.01 Euro/MWh.
Today, the same blocks of (delivery) hours as in the block contract market can be
traded by bidding for the specific hours in the day-ahead market. The orders for
these synthetic blocks can be made in the form of a block bid already discussed
above. However, the block bids do not have a pricing of their own as block contracts
in the block contract market had.
210
Cf. section 4.4.2.1 for a discussion of the daily seasonality in electricity markets.
211
See section 4.4.2.1 for a discussion of negative prices in the day-ahead and intraday market.
212
The weekend base load block contract was introduced at a later date compared to the two
other block contracts, starting trading on November 1, 2002. However, replicating the weekend base
load block contract had already been possible by taking long positions in base load contracts with
delivery on a Saturday and on a Sunday.
61
Futures Market
Together with the options market the futures market forms the derivatives market
of the EEX. Four kinds of futures are traded on the futures market which are char-
acterized by their delivery period, e.g. one week, one month, one quarter, and one
year. The week futures were introduced recently, while the other three futures are
traded from the beginning.
The minimum price fluctuation in the futures market is 0.01 Euro/MWh. Negative
prices are not allowed and a maximum price is not specified. The contract size of
the futures is 1 MW.
The settlement of the traded futures can take place either in cash – these are the so
called Phelix-Futures – or through physical – these are the so called German-Power-
Futures – delivery.213 The week future is only traded with financial settlement.
The overwhelming part of the liquidity in the futures market is observed in the
cash-settled futures. The liquidity in the separate market segments will be further
discussed in section 4.4.1.
There is a base and a peak load version of every future. A base contract ensures
delivery around the clock, independent of working or non-working days. A peak
contract ensures delivery between 8 am and 8 pm and between Mondays and Fridays,
independent of non-working days. A month future, for example, ensures the delivery
of electricity in all hours of any delivery day of a calendar month (base version) or
on all delivery days from Monday until Friday from 8 am to 8 pm (peak version).
The Phelix Base and the Phelix Peak are the underlying for the cash-settled base
and peak futures, respectively. Recently, the product range was extended through
the introduction of an off-peak future. The off-peak future is only offered as a future
with financial settlement, i.e. as a Phelix Off-Peak future. This future is settled
in between midnight and 8 am and between 8 pm and midnight, from Monday to
Friday and around the clock for Saturday and Sunday.
Currently traded delivery periods are the actual week and month, the next four
weeks, the next nine months, the next eleven quarters, and the next six years.214 A
special feature of the futures market is the cascading of the quarter and year futures.
In the case of quarter futures the original future is replaced by three month futures
– together representing the delivery quarter – before the delivery period. The year
future is replaced by three quarter and three month futures. Another interesting
aspect of the futures market is the trading of the month futures in the delivery
213
The futures with physical delivery in France are called French Power Futures. There are no
financially settled French futures.
214
As already discussed above, the number of traded contracts continuously increases as the EEX
extends its product portfolio.
62
month.215
Options Market
Options called Phelix options are traded in the options markets of the EEX. The
Phelix options are European styled options on the Phelix Base future, i.e. the exercise
of these options opens a position in the underlying future. Both puts and calls are
traded. Every option is characterized by the underlying future, the exercise price,
and the maturity. For every underlying future at least three options may be traded,
i.e. at least three different exercise prices are available. The management board of
the EEX may also establish additional option series.
Traded options are available on the respective next five Phelix Base month futures,
the respective next six Phelix Base quarter futures and the respective next three
Phelix Base year futures. The minimum price fluctuation in the options market is
0.001 Euro/MWh. Negative prices are not allowed and an upper price limit does not
exist.
The exercise of an option that is in-the-money takes place on the last trading day.
The exercise of the option opens a position in the underlying future at the exercise
price of the option.
2.2.4.3 Trading Process
Similar to the previous section I separately discuss the trading process in the various
electricity market segments in the following.
Day-Ahead Market
The day-ahead market is operated as an auction market. Thus, all buy and sale
orders are collected in an order book before the auction takes place. For a specific
hour contract orders can be entered, changed, deleted, or retrieved up to 14 days
before delivery. The order book remains the whole time closed, i.e. it is not possible
for the market participants to see the already entered orders. The auction takes
place on the last day before the delivery day of the hour contract.216
Figure 2.11 depicts the detailed trading process on the auction day with a further
specification of the four trading phases distinguished by the EEX for this day. These
phases are the pre trading, the main trading, the post trading, and the batch pro-
cessing phase.
215
Cf. section 4.4.1.2 for a discussion of the trading of the month futures in the delivery month.
216
Cf. EEX (2008b) for further details regarding the trading process.
63
Figure 2.11
Trading Phases Day-Ahead Market EEX
Source: Taken from EEX (2008b), p. 13.
The market participants – both electricity seller and buyer – have two possibilities
to enter hourly bids: the price-dependent and the price-independent bid. The price-
dependent bid consists of the specification of various price-volume combinations.
The minimum and maximum prices are predetermined by the allowed price range.
At least two price-volume combinations at the minimum and at the maximum price
must be specified. Up to 248 more price-volume combinations can be entered. These
are then interpolated to a bidding curve. The price-dependent bid is hence a bid
with continuous price quotations between the minimum and the maximum price.217
These bid curves specify which volume a market participant is willing to buy or to
sell at a certain price. It is common for a market participant to act both as a buyer
and as a seller in the same hour contract in dependence of the price.218 A price-
independent bid on the other side only consists in the specification of a volume that
is to sell or to buy. The bid is executed independent of the price.
The prices for all hour contracts are established in a joint auction. For contracts
with delivery in the German and Austrian market area the auction starts at 12 am.
In the auction all hourly bids are aggregated to form a supply and a demand curve.
An iteration process taking into account various restrictions establishes the prices
for the 24 hour contracts.219 One price is established for every hour contract; every
market participant pays or receives the same price. The auction is hence called a
217
Cf. EEX (2008b), p. 17.
218
Cf. section 2.2.4.4 for a discussion of the market participants and their bidding behavior.
219
Cf. EEX (2008b) for a further discussion.
64
uniform price auction in contrast to a pay-as-bid auction.220 The auction results –

the established price and the traded volume for every hour contract – are published
by the EEX between 12.35 pm and 12.45 pm.
Another characteristic of the trading process in the day-ahead market is the fact that
besides the volume to sell or to buy at a certain price the market participants also
need to specify the control area in which the electricity delivery will take place or
rather is desired.221 In the case of a system congestion between two or more control
areas different prices for the same delivery hour in dependence of the control area can
be established. However, until now congestion did not occur in the German market.
To provide the market participants with further information regarding the auction
results, the EEX publishes the aggregated supply and demand curves of every hourly
auction on the next exchange day at 9 am. Market participants hence have the time
to process the provided information and to adjust their bidding behavior for the
current day.
Intraday Market
The intraday market is operated as a continuous market where matching orders are
executed automatically. Trading in a specific hour contract starts at 3 pm of the
previous day and lasts up to 75 minutes before the beginning of the delivery hour.
The length of the trading period of the hour contract for a certain delivery day is
different since every contract has the same starting point of trading but a different
ending point.
The trading in the intraday market takes place around the clock, seven days a week.
The price formation is based on the immediate execution of matching orders.

The trading in the block contract market included an auction mechanism as well as a
continuous trading period. The trading window in this market segment was between
8 am and 12 am.
The trading process in the block contract market on the trading day consisted in ac-
cordance to the day-ahead market of four phases: The pre-trading, the main-trading,
the post trading, and the batch-processing. During the pre-trading – lasting from
7.30 am to 8 am – market participants could enter, change, delete, and retrieve or-
ders. The order book remained closed during that trading phase. At the beginning
220
Cf. Ockenfels et al. (2008) for a discussion of the two auction forms.
221
From a physical point of view it is not possible to transmit electricity to a certain point. Thus,
the place of delivery is a specified area.
65
of the main-trading phase these orders were then used for the opening auction.222
Afterwards, a continuous trading period followed. At the end of this period again
an auction, the closing auction took place. The orders for the closing auction could
be entered within a five minute timeframe. The post trading and batch-processing
phases served the administration of the trades, the compilation of reports and the
storage of the accrued data.
Futures Market
The futures market is operated as a continuous market. Trading takes place on
exchange days between 8.25 am and 4 pm. The clearing of OTC trades starts at
8.25 am and ends at 5.30 pm.
The last trading day of the year and of the quarter futures is three exchange days
before the commencement of the delivery period. The month futures are traded for
the last time on the day before the last delivery day (Phelix futures) or two days
before the last delivery day. This implies that the month futures are traded during
their delivery month. A special feature of the pricing mechanism in the futures
market is the establishment of daily prices for every tradable future contract.223
This means that even on days when no trading in a certain contract takes place a
price for this future contract is established.224
In this case the price finding mechanism is either based on the order book situation or
on the so-called chief trader procedure.225 Within the chief trader procedure every
market participant is asked by the EEX for a price indication for the non-traded
contracts. The settlement price is then calculated by the EEX as a simple average
price under consideration of certain constraints.
One of the constraints is an existing arbitrage relation between futures with different
delivery periods. To clarify the arbitrage relation a closer look at the quarter future
with delivery in the second quarter of 2009 is taken. For this quarter, five simultane-
ously traded futures with delivery taking place in this quarter exist: the year future
with delivery in 2009, the quarter future with delivery in the specific quarter, and
the three month futures with delivery in April, May, and June 2009.
For the price of the quarter future with delivery in the second quarter in 2009, PQ ,
222
According to Ronn & Wimschulte (2009) almost all trading in the block contract market took
place in the continuous trading period.
223
This is done “for the purpose of the execution of all clearing processes, in particular for the
calculation of the variation margin for every trading participant”; EEX (2008b), p. 50.
224
Cf. section 4.4.1.2 for a discussion of the liquidity in the futures market.
225
Cf. EEX (2008b), pp. 51-52.
66
the following relation applies
1
PQ = (n1 PM 1 + n2 PM 2 + n3 PM 3 ). (2.10)
n
PM 1 is here the price of the month future with delivery in the first month of the
delivery quarter, namely the month future with delivery in April 2009. PM 2 and
PM 3 are the prices of the month futures with delivery in May 2009 and June 2009,
respectively. n1 is the number of delivery days in April 2009, n2 and n3 the numbers
of delivery days in May and June 2010, respectively. n is calculated as the sum of n1 ,
n2 , and n3 . The arbitrage relation arises from the fact that a long position in these
three month futures would be equivalent to a long position in the quarter future.
Traders can earn a risk-less profit when the condition stated in equation (2.10) is
not fulfilled.226
The above equation links the price of the quarter futures to the prices of the month
futures. Thus, if no trading in the quarter futures takes place, the arbitrage-free
price of this future can be calculated with equation (2.10). Needless to say, in this
case trading in the corresponding month futures must take place to establish a reli-
able price. Similar arbitrage relation applies to the other futures. A long position in
a year future, for example, can be replicated by long positions in the corresponding
three month futures and the corresponding three quarter futures. Thus, the price
of the quarter futures discussed above can also be linked to the year future and the
other quarter futures of the corresponding delivery year by an arbitrage relation.
Options Market
The trading hours in the options market are identical to those in the futures market.
The options market is also operated as a continuous market.
The last trading days of the option are: Four days before the beginning of the
delivery period for the Phelix month and quarter options, the second Thursday in
December for the Phelix year option. For options with the delivery period being
the first quarter of a year and January, the last trading takes place on the third
Thursday in December of the previous year. Similar to the futures market prices for
all options are established every day. Non-traded options are valued by an options
pricing model.227
226
Cf. Wimschulte (2010) for a discussion of a similar short-term condition in the Nord Pool and
empirical evidence that the condition is fulfilled, i.e. that the market is efficient.
227
Cf. EEX (2008b), p. 52.
67
Figure 2.12
Market Participants EEX by Number and Country of Origin in 2009
Source: Taken from EEX (2010b), p. 1.
2.2.4.4 Market Participants
As of December 31, 2009 the number of the market participants amounted to 191
in the derivatives market and to 182 in the spot market. Regarding their countries
of origin the distribution of the market participants is found in figure 2.12. The
figure reveals that after Germany the main countries of origin are Great Britain,
Switzerland, Austria, and France.
Market participants on the supply side are mainly utilities, industrial companies with
generation capacities, and foreign electricity importers. On the demand side mainly
industrial companies, supply companies as municipal energy suppliers, trading com-
panies, the trading division of the utilities, banks and other financial institutions are
found.
However, the classification of a market participant to be either a (pure) supplier or
buyer is often not possible. As prior discussed in this section, regarding the trading
process in the day-ahead market, market participants enter hourly bids which place
them on different market sides in dependence of the price. Thus, it is to observe
that market participants regularly change the market side. For a utility with a
peak load power plant, for example, the buying of electricity at the spot market
68
instead of operating its own power plant can be reasonable at certain prices. Thus, a
classification based on the net positions of the market participants in certain trading
intervals is the best classification available mechanism .
The EEX tries to enhance transparency by publishing additional information regard-
ing the trading participation. Considering the day-ahead market the following data
are published on a daily basis:228
• Number of active market participants,
• Number of sellers,
• Number of buyers ,
• Number of net-sellers,
• Number of net-buyers, and
• Average share of the five market participants with the highest revenue (per
market participants).
The data publication for a certain delivery day takes place on the homepage of the
EEX at 9 am the same day.
In addition, the following data considering the trading in the futures market are
published on the first exchange day of a calendar month:229
market participants) in the Phelix month futures,
market participants) in the Phelix month futures including the OTC market,
and
• Share of the market makers on the total revenue in the derivatives market.
According to the EEX the number of net-sellers in 2008 was on average higher than
40, more than half of these sellers not being from Germany.230
A further measure to increase market transparency is the publication of generation
and consumption data on a neutral transparency platform.231 Around 80% of the
German capacity is covered by this platform.
228
Cf. EEX (2009a), pp. 64-65.
229
Cf. EEX (2009a), p. 65.
230
Cf. EEX (2009a), p. 59.
231
Cf. the website of the transparency platform (www.transparency.eex.com) for a discussion of
the provided data.
69
Figure 2.13
Evolution of German Wholesale Electricity Price 2002-2010
80.00
70.00
60.00
50.00
Price [Euro/Mwh]
40.00
30.00
20.00
10.00
0.00
2002 2003 2004 2005 2006 2007 2008 2009 2010
Source: Own work, based on data from EEX.
2.2.4.5 Wholesale Electricity Price
The average (daily) hourly wholesale electricity price in the German day-ahead mar-
ket amounted to (38.89) 38.86 Euro/MWh in 2009. In 2003, the first full year of
existence of the EEX, the average price was (29.73) 29.49 Euro/MWh. This price
evolution would correspond to a price increase of approximately (31) 32% over six
years. However, a near analysis of the time series reveals that over the first six years
a gradually price increase took place. The time series is shown in figure 2.13. It can
be seen that in 2008 a price peak was reached. The average price in 2008 amounted
to 66.14 Euro/MWh. In addition, a large price increase is observed in 2005. Overall,
an increasing trend is observed over the first years. This trend in the price evolution
was stopped by the recent financial and economic crisis which caused a strong price
decline in 2009.
A continuously increasing electricity price over the whole lifespan of the EEX seems
to be a characteristic of the liberalized German electricity market. The literature
mainly identifies two fundamental drivers of this development: the increasing fuel
costs, in particular the anthracite price and gas price, and the introduction of the
emission rights trading in 2005.232
232
Cf. for example Schwarz et al. (2007) and Bundesministerium für Wirtschaft und Technologie
70
In particular the introduction of the European Union Emission Trading System (EU
ETS) has marked a significant change for the established market structures since
the liberalization. A completely new cost block was introduced through this change,
namely the exchange traded emission right or, officially, the emission allowance unit
(EAU).233 The EAU grants the right to emit one tonne carbon dioxide (CO2 ) or an
equivalent in the European Union. At the beginning of the trading in 2005 the EAUs
were allocated for free; a gradual increase of the sold emissions is intended over the
years. However, opportunity cost considerations resulted already in the first year of
the EU ETS in an sharp increase of the wholesale electricity price as the electricity
generation side incorporated the emission right’s price (at least partially) into the
electricity price. The results were high windfall profits for the power sector234 . Since
then, the electricity price development is directly linked to the emission rights price
development.235 However, Zachmann & von Hirschhausen (2008) show in an early
work that the price changes in the emission rights market are passed asymmetric into
the electricity prices. According to the results of the authors positive price changes of
emission rights have a stronger impact on the wholesale electricity price than falling
prices.
In addition, parts of the academic literature show that besides these factors market
power influences (or influenced in past periods) the electricity prices as well236 . Mar-
ket power is, in general, defined as the ability (of market participants) to profitably
alter prices away from competitive levels (fair market prices); market power is widely
discussed in the literature. The discussion on the existence of market power in the
German electricity market and on measures to migrate it is almost as old as the
EEX.237 The methodology applied in academic studies consists almost always in the
ex post comparison of observed and modeled electricity prices and is very sensitive to
the underlying assumptions regarding the modeling of the theoretical prices.238 Un-
surprisingly a reasonable number of studies finds mixed or even contrary results.239
(2009).
233
The EU ETS implements a cap-and-trade mechanism with an absolute (emission) cap within
the European Union; cf. for example Benz & Trück (2006), Convery & Redmond (2007), Ellerman &
Buchner (2007), Kruger et al. (2007), Daskalakis & Markellos (2008), and Daskalakis et al. (2009)
for details regarding the emissions trading in the European Union and the implemented trading
mechanism.
234
Cf. for example Frondel et al. (2008).
235
Cf. for example Sijm et al. (2006) for a discussion of the interdependence between the emissions
rights price and the electricity price.
236
Cf. for example Müsgens (2006), von Hirschhausen et al. (2007), Weigt & von Hirschhausen
(2008), and Janssen & Wobben (2009).
237
Cf. Schwarz & Lang (2006) for one of the first empirical studies on market power in the German
wholesale market.
238
Cf. Ellersdorfer et al. (2008) for a discussion of potential problems when modeling theoretical
prices.
239
Cf. for example Schwarz et al. (2007) and Möst & Genoese (2009)
71
The results for other markets are mixed as well.240 Further problems complicate the
detection of market power even more.241
240
Cf. Fridolfsson & Tangera (2009) for a discussion of market power in the Nord Pool.
241
Cf. Bautista et al. (2007) for a discussion of these problems.
72
2.3 Price Formation in Commodity and Electricity Fu-

tures Markets
The research on price formation in commodity futures markets dates back to the
work of Keynes and his theory of backwardation. Two theories have been established
since then in the academic literature: the theory of storage and the risk premia ap-
proach. As the empirical literature finds support for both theories an understanding
of the similarities and differences between these two theories is necessary in order to
evaluate their suitability in the case of electricity markets.
In this section I summarize the theoretical and empirical research on price forma-
tion in commodity futures markets. Thereby, I focus on commodity futures markets,
as the main characteristic that is of interest in the case of electricity is storability.
After an introduction to the topic I provide an overview on the theory of storage
and the risk premia approach. I discuss the theoretical background of these theo-
ries and then address the specifics of electricity futures markets, in particular the
non-storability from an economic point of view and the resulting futures character
of all exchange-traded electricity contracts. Afterwards, I report the empirical evi-
dence on price formation in commodity futures markets over the last three decades.
Furthermore, I devote considerable attention to a literature review on the empirical
research on electricity futures markets. Thereby, I distinguish between theoretical
results, empirical results on short-term contracts, and empirical results on long-term
contracts. Short-term contracts are generally defined as day-ahead contracts, while
long-term risk contracts are defined as week and month futures. I summarize the
obtained results according to magnitude and the sign of risk premia. Moreover, I
address time variation, seasonality and their potential drivers.
• Which are the standard theories for price formation in commodity futures
markets?
• Which of these price formation theories is appropriate for electricity futures

markets?
• What is the empirical evidence on price formation in electricity futures mar-

kets?
73
2.3.1 Introductory Remarks
Shortly after the beginning of the worldwide deregulation of electricity markets and
the establishment of electricity exchanges as wholesale markets, the question on the
theoretical foundation of price formation in electricity (futures) markets was posed.
In contrast to financial and other commodities markets, where the theory of storage
as a non-arbitrage condition can be mostly applied, electricity is non-storable. As
storability is the basic assumption of the theory of storage, this mechanism is not
applicable242 in the case of electricity futures.243 Thus, the question about the
mechanism behind price formation in electricity futures markets has accounted for
immense research over recent years.
From an equilibrium point of view, the risk premia approach seems to be the most
promising price formation mechanism. In general, this approach identifies two possi-
ble determinants of risk premia: systematic risk and hedging pressure (Bessembinder
(1992)). The existence of the first determinant, systematic risk, is under controversial
discussion. The main part of the empirical literature finds no evidence to support
a systematic risk component in futures returns. The second determinant, hedg-
ing pressure, was introduced with the normal backwardation theory formulated by
Keynes (1930) to the academic literature. Later, this theory was extended to the
general hedging pressure theory, which indicates that the price formation of futures
is based on two components, the expected spot price at maturity of the future and
a risk premium. The risk premium is paid by risk-averse market participants as a
compensation for the elimination of price risk.
There are currently mixed empirical results regarding the hedging pressure theory
as an appropriate price formation mechanism in commodity markets.244 Empirical
literature suggests that this theory is appropriate for certain commodities, mostly
characterized through no or rare storage possibilities. The work of Fama & French
(1987) is one of the classic empirical studies on this topic.
Regarding price formation in electricity futures markets, the hedging pressure the-
ory seems to be an appropriate approach.245 However, the empirical results raise
questions on the magnitude and sign of the risk premia. In the next chapter I focus
on these questions for the German market.
242
Cf. Schwarz et al. (2007), pp. 58-59.
243
However, Botterud et al. (2010) argue that in electricity markets with a high share of hydro
power the theory of storage may also be relevant, as hydro power implies storability to a certain
degree.
244
Cf. section 2.3.3.1 for a discussion of the empirical results on commodity markets.
245
Cf. section 2.3.3.2 for a discussion of the empirical results on electricity markets.
74
2.3.2 Price Formation in Commodity Futures Markets: Theory
2.3.2.1 Theoretical Approaches
Two standard theories are discussed in the academic literature as potential foun-
dations for price formation in commodity futures markets246 : the theory of storage
and the risk premia approach.247 The theory of storage links futures and spot prices
through a no-arbitrage condition and relies on the storability of the underlying com-
modity.248 The risk premia approach is a general equilibrium theory and links futures
and spot prices through expectations regarding the future spot price and a risk pre-
mium. Two potential determinants of the risk premium, namely systematic risk and
hedging pressure, are identified. Therefore, the term risk premia approach is not
well-defined.
Theory of Storage
The theory of storage, also known as cost-of-carry approach or hypothesis, is based
on arbitrage considerations and relies on the assumption of storability of the under-
lying commodity. Kaldor (1939) and Working (1949) originated the theory. Later,
Brennan (1958) and Telser (1958) refined and formulated the theory as it is known
today.
Three factors, two of them fundamental in nature, form the backbone of the theory.
The first fundamental factor is the interest that could be earned by the long position
in a future on the fixed futures price. The second fundamental factor is the storage
cost which occurs for the physical holding of the underlying asset. The third factor is
generally called convenience yield and expresses the benefit from physically holding
the asset.
In its analytic form the theory of storage expresses the futures price F(t, T) at time
t with maturity in T as
F (t, T ) = S(t)er+u−y (2.11)
with S(t) as the spot price at time t, r the (risk-free) interest rate, u the storage
cost, and y the convenience yield.
246
Regarding the pricing of financial futures, Copeland et al. (2005) state that “Financial instru-
ments are usually traded in very liquid spot markets, and there is virtually no cost of storage. This
distinguishes them from commodity markets where the spot market may be thin and storage costs
high. It also makes financial futures somewhat easier to price because arbitrage between the spot
and futures markets helps to determine the future price.”; Copeland et al. (2005), p. 286.
247
Cf. Chow et al. (2000) for a recent survey on price formation in futures markets.
248
In the following I consequently use the term commodity. However, most of the argumentation
is applicable to other assets as well.
75
The two fundamental factors in equation (2.1), the interest rate and the storage
cost, are in reality always positive. The result is a futures price that is higher than
the spot price. The convenience yield can be both positive and negative. Thus,
depending on the market conditions this factor may lead to a higher or lower futures
price compared to the spot price. The theory of storage can hence depict both a
market in contango and a market in backwardation.249
The convenience yield is the crucial factor of the theory of storage as it is not directly
observable. The convenience yield reflects the benefit of a physical holding of the
commodity, i.e. the advantage of effectively possessing the commodity instead of
having to rely on a (liquid) market. Kaldor (1939) states that “stocks of goods
... have a yield, qua stocks, by enabling the producer to lay hands on them the
moment they are wanted and thus saving the cost and trouble of ordering frequent
deliveries, or of waiting for deliveries”.250 The perception that the convenience yield
is reciprocally proportional to the stocks of the underlying is the basic assumption
in academic literature.
In an empirical verification of the theory of storage, the convenience yield is, in gen-
eral, a residual factor. It is approximated as reciprocally proportional to the stocks
of inventory. However, the approximation is still rough because of to two facts: first,
the approximation by stocks of inventory itself is a strong simplification, and second,
it is obviously difficult to estimate the worldwide stocks of an inventory. Similar to
the implicit volatility in the Black-Scholes(-Merton) model, empirical research faces
the problem that it can hardly falsify the model through an ex post analysis.
Risk Premia Approach

The risk premia approach identifies two potential determinants of risk premia: sys-
tematic risk and hedging pressure. In the following, I discuss both determinants
individually.
Systematic Risk
The first step in valuating a financial instrument is, in general, the application of
the CAPM. The intention is to determine a risk premium from the systematic risk
in the returns of an asset. The earliest academic contributions on the valuation
of commodity futures applying the CAPM are found in Dusak (1973) and Black
(1976a).
Before a discussion on the valuation of commodity futures applying the CAPM can
249
A market in contango is characterized by futures prices being higher than the spot price; a
market in backwardation is characterized by futures prices lower than the spot price.
250
Kaldor (1939), pp. 3-4.
76
take place, a question which is essential for the application of this approach needs
to be posed. The question is whether a commodity future may be treated as an
asset, a necessary condition for applying the CAPM. Black (1976a) is of the opinion
that commodity futures do not belong to the market portfolio. The author justifies
this by stating that for every long position there is one short position – both thus
neutralizing each other. This implies that commodity futures can be seen as perfect
zero-sum games. In other words, according to Black (1976a), these futures contracts
are mere bets on future prices of a good or a commodity. Dusak (1973), on the
other side, argues contrarily, seeing futures markets as normal capital markets where
trading takes place in the same way as with any other asset.
Following the argumentation that the valuation of futures based on the CAPM is
appropriate, it is possible to calculate the futures price as the expected spot price
adjusted for the expected risk premium.251 In a one-period framework, it is possible
to note the futures price as252
F (t, T ) = Et [S(T )] − [E(Rm − Rf )]St β. (2.12)
The futures price F(t, T) is equal to the expected spot price, Et [S(T )], minus a
risk premium. This risk premium corresponds to the expected return from taking
a long position in a future. The risk premium is defined as the sum of the risk-free
interest rate, Rf , and of a component which is proportional to the systematic risk of
the given future. The systematic risk, β, is defined as the covariance of the futures’
returns with the returns of the market portfolio, Rm . It reflects the non-diversifiable
share of the total risk. St is the spot price at time t.
As the covariance between the futures returns and the returns of the market port-
folio can be either positive or negative, this approach can model both a market in
contango or a market in backwardation.
Hedging Pressure
The hedging pressure theory arises from equilibrium considerations and dates back
to Keynes (1930) and Hicks (1939).253 Later that approach was generalized to the
hedging pressure theory (Cootner (1960)). More recently, systematic risk and hedg-
ing pressure have been merged to joint models.254
Keynes (1930) assumes in his original work that producers always pay the risk pre-
251
Cf. Dusak (1973) for the derivation of the analytical relation and potential problems.
252
Cf. Copeland et al. (2005), p. 291.
253
John M. Keynes had formulated his idea on the normal backwardation for the first time already
in an essay in the Manchester Guardian Commercial in 1923; cf. Keynes (1923).
254
Cf. for example Stoll (1979), Hirshleifer (1988), and Hirshleifer (1989).
77
mium in order to get rid of their price risk.255 From Keynes’ perspective, in the
words of Dusak (1973), “a futures market is an insurance scheme in which the specu-
lators underwrite the risks of price fluctuation of the spot commodity”.256 Under the
assumption that the expected spot price equals the current spot price, this results in
a downward sloping term structure, i.e. a market in backwardation. Thus, Keynes’
theory is termed the normal backwardation theory. The generalization was derived
from the insight that consumers can pay the risk premium as well.
The hedging pressure theory is based on the expectation theory.257 The hypothesis of
the unbiased estimator is dropped and the existence of a risk premium postulated.
The futures price is expressed as the sum of the expected spot price and a risk
premium. Thus, in general, the futures price is a biased estimator of the spot price.
The hedging pressure theory assumes that futures markets consist of two general
types of market participants: hedgers and speculators.258 Market participants act-
ing as hedgers are assumed to have a real interest in the underlying of the future.
Their motivation, for example, is to ensure a smooth production process through
hedging of the sales prices or actually delivering the commodity in the future. Mar-
ket participants acting as speculators have only a financial interest.
Market participants seeking financial gains only enter a financial market if they can
expect, on average, a positive return. Thus, speculating market participants need to
be offered a positive expected return for taking of a position in the futures market.
Thus, the offered bearing of the price risk must be compensated via a risk premium.
As hedgers may go short as well as long in the futures markets, it has become common
to speak of producers and consumers. Normally, a producer possesses the commodity
and takes a short position in a future in order to hedge the future price. On the other
side, a consumer may have a need for the future’s underlying and takes a long position
in the specific future to secure the future price. Depending on the market structure,
on the macroeconomic framework, and on various other factors, there may be a
predominance of either producers or consumers acting as hedgers. Needing to hedge
the future prices, consumers are willing to pay a premium for the counterparty’s
255
The original theory of normal backwardation as formulated by Keynes (1930) is based on three
main assumptions (Chatrath et al. (1997)): (1) market participants acting as speculators are risk
averse (2) market participants acting as hedgers are net short, and (3) market participants acting as
speculators do not have forecasting ability. The second assumption was released by Cootner (1960).
The third assumption is discussed in more detail in section 4.3.
256
Dusak (1973), p. 1387.
257
In the financial literature the term expectation theory is generally used for a theory of the yield
curve and forward interest rates. According to this theory, market participants’ decisions on capital
allocation are influenced by their expectations towards the future interest level. This implies that,
according to the pure expectation hypothesis, futures interest rates and forward interest rates, may
be interpreted as the unbiased estimators of future spot interest rates.
258
Some authors see arbitragers as a further type of market participants; cf. for example Geman
(2005).
78
willingness to take the counter position. The price risk as well as the corresponding
risk premium are then transferred from the hedger to the counterparty in the future.
The counterparty can be a hedger or a speculator. If there are as many producers
as consumers neither of the parties needs to pay a risk premium; speculators are not
required. Risk premia exist only if there is a predominance on one side (leading to
the term hedging pressure), i.e. a net hedging pressure. Speculators are motivated
to enter the market and take the counter position in the future in order to gain the
risk premium.
The futures price F(t, T) at time t with maturity in T is defined by the risk premia
approach in a formal way as
F (t, T ) = Et [S(T )] + π(t, T ) (2.13)
with Et [S(T )] being the expected spot price at maturity T and π(t, T ) the risk
premium.
According to the hedging pressure theory, depending on the distribution of market
participants, the futures price can be lower or higher compared to the spot price.
Therefore, the theory is able to describe a futures market in contango as well as in
backwardation.
2.3.2.2 Specifics of Electricity Futures Markets
The academic literature assumes that the risk premia approach is the appropriate
price formation mechanism for exchange-traded electricity contracts. This point of
view is mainly based on the non-storability of the commodity electricity and the
resulting futures market character of all electricity markets. Therefore, the price of
an electricity contract is mostly seen to include a risk premium in addition to the
forecast of the future spot price.
As the theory of storage is based on the assumption of storability, it does not have to
be applicable in the case of electricity. Furthermore, as noticed by Geman & Vasicek
(2001), “the non-storability of power makes irrelevant (...) the notion of convenience
yield, which represents the benefits accrued from ’holding’ the commodity”.259
The risk premia in electricity contracts can be both positive and negative. Empirical
results indicate that risk premia in electricity markets are mostly positive, at least
for short- and mid-term futures.260 This is contrary to other markets as positive risk
259
Geman & Vasicek (2001), p. 93.
260
The terminology used in this dissertation is as follows: Futures with a maturity between one
and three months are considered to be short-term, with a maturity between four and twelve months
as mid-term, and with a maturity over twelve months as long-term futures.
79
premia translate into a negative price of risk.261 Thus, a long position in these con-
tracts is on average linked to negative returns. A plausible economic interpretation
of positive risk premia is that holders of long positions in the futures are compen-
sating holders of short positions for bearing the price risk. Under the assumption
that prices are set by industry participants – and not by outside speculators – this
implies that electricity consumers are more interested in hedging than the produc-
ers are. This explanation seems to be appropriate considering that price risk is an
essential risk in the short run, mainly due to frequently emerging price spikes.
Assuming that electricity consumers are mainly interested in hedging their short-
term price exposure, one can argue that the sign of the risk premium can change
according to the time horizon considered. Electricity consumers use short-term fu-
tures for hedging purposes while producers predominantly use long-term futures.
The economic rationale behind the producer behavior may be the long-term char-
acter of investments in the energy industry. This results in demand for long-term
futures to hedge cash flows far in the future to gain at least some planning relia-
bility for investment decisions.262 In consequence, the behavior of both consumers
and producers may result in market segmentation which translates into positive risk
premia in short-term and negative risk premia in long-term futures. Benth et al.
(2008) develop a framework to model this effect.263
However, parts of the academic literature disagree with the interpretation of price
differences between electricity markets or rather market segments with different trad-
ing times as risk premia. Borenstein et al. (2008), for example, argue that market
inefficiencies and market power are alternative explanations for these price differ-
ences.264 Factors such as luck or superior forecasting abilities265 may also serve as
explanations for these price differences.
Another interpretation of the price differences are forecast errors. However, in the
case of longer sample periods, the occurrence of systematic forecast errors poses the
question why arbitrageurs do not profit on these errors or rather whether expecta-
tions are formed rationally. To apply the ex post risk premia approach266 , one of
the underlying assumptions is rational expectations; systematic forecast errors are
excluded under this assumption.
261
Cf. Kolos & Ronn (2008), p. 623.
262
Cf. chapter 3 for a discussion of the impact of hedging on the NPV of the investment.
263
Cf. section 2.3.3.2 for a discussion of the model developed by Benth et al. (2008).
264
Borenstein et al. (2008) also discuss transaction costs as potential reason for persistent price
differences. However, the authors observe that at least direct transaction costs are too small in
magnitude to be used as an explanation.
265
Cf. section 4.3 for a discussion of the impact of superior forecasting abilities on returns in
futures markets.
266
Cf. section 4.3 for a discussion of the methodology applied in this dissertation and the under-
lying assumptions.
80
2.3.3 Price Formation in Commodity Futures Markets: Evidence
2.3.3.1 Empirical Evidence Commodity Futures Markets
A broad literature on the suitability of the risk premia approach and of the theory
of storage as a potential price formation mechanism in futures markets exists.267 I
focus the literature review on empirical studies dealing with commodity markets and
restrict the review to the past 30 years. Furthermore, I restrict the literature review
on the classic studies. The empirical results on the suitability of the risk premia
approach (or rather the normal backwardation theory as this is the term most often
used in this context) are of primary interest for the further discussion.
Bodie & Rosansky (1980) compare returns of commodity futures with returns on
common stocks. Using a sample consisting of 23 commodities traded in the United
States over the period from 1950 until 1976 the authors find a positive mean rate of
return of a benchmark portfolio. The return of the portfolio is found to be about
the same as the mean rate of return on common stocks. When estimating the beta
of their commodity portfolio the authors reject the hypotheses of the beta being
equal to one at any significance level. Regarding the betas of the individual futures
the most are found to have a negative beta over the sample period. Furthermore,
the authors show that commodity futures seem to be a good inflation hedge. Their
results indicate that stocks and futures are negatively correlated. In summary, the
authors see their results as support for the normal backwardation hypothesis.
The results of Bodie & Rosansky (1980) also enlarge the empirical results of Dusak
(1973). When analyzing the average returns for wheat, corn, and soybean futures
over the period 1952 to 1967 Dusak (1973) finds results close to zero. By enlarging
the sample period, Bodie & Rosansky (1980) find evidence on positive returns for
these commodities.
Soybean, corn, wheat, cotton and cattle futures markets are investigated by Carter
et al. (1983). The authors analyze weekly average futures prices over the period 1966
to 1976 and aim to evaluate the portfolio interpretation of futures. They extend the
framework of Dusak (1973) and find some evidence for systematic risk. In addition,
they find evidence for non-market risks with a seasonal pattern. The betas are within
their framework stochastic and the market portfolio includes futures contracts.
Jagannathan (1985) aims to investigate commodity futures prices using the consumption-
based intertemporal CAPM. The developed framework is tested on futures prices for
corn, wheat, and soybeans over the period January 1960 to December 1978. The
empirical evidence result in a rejection of the framework.
267
Cf. for example Carter (1999) for a recent literature survey.
81
The profitability of certain trading strategies in three commodity markets (wheat,

corn, and soybean) is tested by Chang (1985). His data span over the period from
1951 to 1980. The obtained results are evidence for the theory of normal backwar-
dation, i.e. the author indirectly finds that a risk premium is exchanged between
hedgers and speculators. Furthermore, the author shows that the risk premia tend to
become more significant in recent years when compared with earlier years and that
certain speculators seem to posses superior forecasting abilities. However, Chang
(1985) makes no attempt to estimate the relative size of the sources of returns.
Hartzmark (1987) rejects the theory of normal backwardation and its extension.
Analyzing trading histories of individual futures traders in nine futures markets the
author finds no consistent (positive) risk premia received by speculative traders.
Thereby, his dataset spans over four years and is taken from a confidential file from
the Commodity Futures Trading Commission (CFTC).
Using the same dataset, Hartzmark (1991) shows in a further paper that individual
traders do not possess superior forecast abilities. The conclusion of the author is
that trading performance is driven by luck.
Fama & French (1987) analyze price formation in commodity futures markets by
examining the suitability of the theory of storage and the risk premium approach.
They analyze 21 commodities (agriculture, wood, animal, and metal products); while
sample periods vary, they mostly analyze data between 1966 and 1984. While finding
support for the first theory, the results for the hedging pressure theory are mixed;
with only marginal evidence found. Thus, the authors conclude that their results
are not strong enough to make a contribution to the controversy on whether nonzero
expected risk premia exist.
Bessembinder & Chan (1992) examine whether instrumental variables like Treasury
bill yields and equity dividend yields, which were shown to possess forecast power
in equity and bond markets, can be applied in futures markets (agriculture, metals,
and currency). They show that the used instrumental variables also possess forecast
power in 12 futures markets. The authors conclude that their results are consistent
with time-varying risk premia.
The roles of systematic risk and hedging pressure in explaining risk premia in futures
are investigated by Bessembinder (1992). Using data from equity markets and 22
financial, foreign currency, agriculture, and metal futures markets, covering mostly
the sample period January 1967 to December 1989, the author finds little evidence
for the first determinant but substantial evidence for hedging pressure effects.
Kolb (1996) focuses on systematic risk and returns in futures markets. The author
examines 4735 futures contracts on 45 commodities with almost 600,000 daily ob-
82
servations. The covered sample period is 1969 to 1992. The author investigates
three key hypotheses: (i) the mean returns are equal to zero, (ii) the systematic
risk of futures contracts is zero, and (iii) the relationship between realized returns
and systematic risk. The results of the empirical analysis are: (i) 29 of the futures
have returns that do not differ significantly from zero, (ii) the mean beta is 0.0463 for
physical commodity futures and only five goods have mean betas exceeding 0.10, and
(iii) there is no positive relationship between realized returns and systematic risk.
If anything, the returns are generally inversely related to systematic risk. However,
due the low levels of systematic risk these results must be interpreted with caution.
Empirical evidence regarding the existence of hedging pressure is reported by de Roon
et al. (2000). The authors analyze 20 futures markets, divided into four groups
(financial, agriculture, mineral, and currency). Their dataset spans over the period
January 1986 to December 1994 and consists of semi-monthly price observations.
Finding evidence for hedging pressure effects, the authors show that the results are
robust to market risk considerations.
Regarding energy markets, Considine & Larson (2001) test for the presence of risk
premia in crude oil and natural gas markets. The authors analyze the futures contract
for West Texas Intermediate crude petroleum over the period 1983 to 1999 and the
natural gas futures contracts for delivery at the Herny Hub over the period 1990 to
1999. They find strong evidence for risk premia and also show that these premia
rose sharply with increasing price volatility. In addition, the authors find evidence
supporting the existence of convenience yields.
Wei & Zhu (2006) aim to estimate the convenience yield and the risk premium in the
natural gas market of the United States. The authors use contrary to other studies
forward prices instead of futures prices. The forward price data cover the period
January 1991 to August 2003. Regarding risk premia the authors conclude that they
are measurable and economically significant. When analyzing the determinants of
the risk premia they find mixed results.
2.3.3.2 Empirical Evidence Electricity Futures Markets
Empirical results regarding the existence of risk premia in electricity contracts can
be divided into two lines of research. The first focuses on short-term risk premia,
mainly defined as the price difference between the hour contracts in the day-ahead
and in the intraday market. The second line examines long-term risk premia, mostly
focusing on the analysis of week and of month futures.
Contributions to the research on short-term risk premia are made among others
by Geman & Vasicek (2001), Saravia (2003), Longstaff & Wang (2004), Boogert &
83
Dupont (2005), Karakatsani & Bunn (2005), Diko et al. (2006), Hadsell & Shawky
(2007), Borenstein et al. (2008), Hadsell (2008), Douglas & Popova (2008) Daskalakis
& Markellos (2009), Ronn & Wimschulte (2009) and Viehmann (2009). The results
of this research are mostly the detection of risk premia which vary throughout the
day and are highly volatile. In general, the risk premia are positive during hours of
high demand.
Shawky et al. (2003), Lucia & Torro (2008), Botterud et al. (2010), Marckhoff &
Wimschulte (2009), Capitan Herraiz & Rodriguez Monroy (2009), Redl et al. (2009),
Furio & Meneu (2010), Kilic & Huisman (2010), Wilkens & Wimschulte (2007)
Bierbrauer et al. (2007), and Kolos & Ronn (2008), among others, contribute to
the research on long-term risk premia. The results are mostly positive risk premia
in the short-term and mixed results for the long-term.
Prior to individually discussing the results obtained by these two lines of research, I
address the most important contributions of the theoretical literature on risk premia
and price formation in electricity futures markets.
Theoretical Literature
The work of Bessembinder & Lemmon (2002) is probably the most influential the-
oretical paper on electricity futures markets, at least according to the number of
citations. The authors develop an equilibrium model for electricity forward pricing
with closed form solutions. They assume that prices are determined by industry
participants and that the mean and the variance of the profits are of interest for
power companies. The authors derive the net demand for forward contracts, both
of the producer and of the retailer. Implications of their model are negative risk
premia in the case of expected low demand and demand risk. An increase of these
two variables leads to an increase of risk premia which can even result in positive
risk premia. Thus, the model links risk premia to risk considerations.
Benth et al. (2008) develop a model that is able to explain the dependence between
the risk premia, the risk preferences and the interaction between buyers and sellers.
Under certain conditions they can derive explicit solutions. The authors argue that
producers are exposed to market risk over longer time periods than consumers. This
has a first order impact on forward prices and the risk premia. Applying this model
to the German market the authors show that the risk premia exhibit a term structure
depending on the risk aversion and the market power of producers.
A model where electricity prices are explained by demand and capacity is developed
by Cartea & Villaplana (2008). The model enables the authors to derive analytical
expressions to price forward contracts and to calculate risk premia. When applying
84
the model to data from the United States, England, Wales, and the Nord Pool the
authors observe that the dynamics of the risk premia are seasonal. This is according
to the observed volatility of demand as the demand follows also a seasonal component.
In months with high volatility of demand, forwards trade above the spot prices; in
months with low volatility of demand even negative risk premia are possible.
Pirrong & Jermakyan (2008) propose and implement a model to valuate power and
weather derivatives. The electricity spot price is in this model a function of two state
variables: demand and fuel price. The application of the model to data from the
PJM market results in upward biased forward prices. Furthermore, the authors show
that the upward bias is higher for forwards with maturity in high demand periods.
The authors explain the large biases with extreme right skewness of power prices
which induces left skewness in the payoff distribution of the short side; a large risk
premium is hence required to keep the seller of power forwards in the market. Based
on their results the authors conclude that the power markets are not fully integrated
with the broader financial markets.
The influence of natural gas storage inventories on the risk premia is investigated
by Douglas & Popova (2008). The authors develop a model that links the effect of
gas storage constraints on the higher moments of the electricity price distribution.
The developed model confirms the results of Bessembinder & Lemmon (2002) and
extends their model. One of the model’s predictions is a negative effect of an increase
in gas storage inventories – under certain conditions – on the risk premia. Testing
the model on data from the PJM market the results strongly support the model. The
analyzed dataset ranges from January 1, 2001 to December 31, 2004 and consists
of hourly prices of the day-ahead and of the intraday market. The estimated risk
premia vary throughout the day.
Support for the results of Douglas & Popova (2008) is found by van Treslong &
Huisman (2009). The authors apply a different definition of the risk premia and find
similar results. Thus, they conclude that the results are not influenced by modeling
effects.
Daskalakis & Markellos (2009) research the topic on whether the risk premia in elec-
tricity forward contracts are affected by emission allowance prices. The trading of
emission allowances within the EU ETS established a new direct production cost
factor for the electricity generator, directly impacting the electricity prices. When
analyzing the impact on the risk premia the authors find results that indicate that
the relation is positive, i.e. they establish a link between the emission allowance spot
returns and the risk premia. To establish this link they first have to obtain results
for the German intraday market. When analyzing the period September 2006 to
October 2007 – approximately the first year of operation of the intraday market –
85
the authors detect negative daily risk premia. However, the results must be inter-
preted as being preliminary due to the short sample period and due to the fact that
liquidity during the first months of trading in the intraday market was very low.268
Furthermore, the authors sample the data at daily frequencies. They also find risk
premia in contracts traded at the Nord Pool and at the Powernext.
Short-term risk premia

Geman & Vasicek (2001) present a formalism for pricing derivatives and modeling
spot price behavior for non-storable commodities. They focus on the commodity
electricity due to the recent introduction of exchange-traded contracts and its unique
characteristics. When analyzing data from the Western Hub of the Pennsylvania,
New Jersey, and Maryland (PJM) market they find positive risk premia. Thereby,
the authors compare 740 day-ahead prices with realized spot prices. Furthermore,
they find evidence for higher risk premia in summer months.
Saravia (2003) analyzes the New York market which is operated by the New York
Independent System Operator (NYISO). This market began operation in November
1999. During the first two years of operation market access was exclusively provided
to market participants with generation capacity or who were responsible for procur-
ing electricity. Then, in November 2001, virtual bidding was implemented, i.e. a
mechanism intended to open the market to non-commercial market participants by
allowing speculating on day-ahead and on intraday market price differences. The au-
thor finds positive price differences, i.e. risk premia, for these two market segments
in the analyzed period. However, Saravia (2003) shows that due to speculators’
participation after the introduction of the virtual bidding the risk premia decreased
substantially. Thus, the results indicate that the increase of market participants
leads to a more efficient market.
One of the first published academic and probably by now the most-cited empirical
paper regarding short-term risk premia is Longstaff & Wang (2004). The authors
analyze the PJM market over the period June 2000 to November 2002. They use
a dataset consisting of hourly spot and day-ahead prices extending over 913 days.
After finding significant risk premia which systematically vary throughout the day
they link these risk premia to different risk factors as volatility of unexpected changes
in demand, spot prices, and total revenue. They find evidence that the risk premia
are positively related to these risk measures. The mean hourly risk premium is
estimated at 0.59 U.S. Dollar, whereby 10 hourly risk premia are negative and 14
268
I exclude the sample period covered by the analysis of Daskalakis & Markellos (2009) in the
empirical analysis in the next chapter as the liquidity of the intraday market was extremely low at
the beginning; cf. section 4.4.1.1.
86
positive. However, the average daily risk premium is statistically not significant.
The highest risk premia are found in the peak evening hours with around 12% of the
average spot price. In general, the largerst risk premia are found during the 12 pm to
9 pm period. The authors also test the implications of the Bessembinder & Lemmon
(2002) model and find support in the analyzed data. Comparing the volatility of
forward and expected prices the authors find a lower volatility of the forward prices.
They interpret this observation as a further evidence for risk premia in the forward
prices. The conclusion of the study is that forward prices in the PJM market are
rationally determined by risk-averse market participants.
Boogert & Dupont (2005) analyze the relation between the intraday and the day-
ahead market in the Netherlands. The primary aim of their paper is to test the
effectiveness of the policy of the Dutch regulator to prevent trading across these two
markets. Trading based on a spread strategy is seen as gambling by the regulator
which therefore implemented a system to prevent this kind of trading. The rationale
is preventing imbalances in the Dutch market. Using a dataset covering the period
January 2001 to December 2003 the authors find that the price difference between the
two markets, i.e. the risk premia, are slightly positive but not statistically significant.
To conduct this analysis the prices from the intraday market which have a frequency
of 15 minutes are aggregated to hourly prices. When testing the profitability of two
basic trading strategies (selling electricity in the day-ahead market and buying it in
the imbalance market and buying electricity in the day-ahead market and selling in
the intraday market) based on these risk premia the authors find weak evidence for
a profitability. Thereby, the profitability is tested over the whole period, over each
single year, and over each single hour.
The focus of Karakatsani & Bunn (2005) is on the day-ahead risk premia in the
British market. The authors use a dataset from June 2001 to June 2003 in order to
verify the existence of risk premia. They classify the half-hourly trading periods in
two homogeneous clusters, peak (7am – 7pm) and off-peak (7pm – 7 am), regarding
them as fairly homogeneous in terms of hedging incentives, demand characteristics,
and operating plant technologies. The authors find significant risk premia which
change sign, depending on whether peak or off-peak hours are analyzed. As poten-
tial explanations for this change in sign they identify the asymmetric positions of
generators and suppliers towards risk and the possibility of an intra-day variation.
The estimated risk premia are positive on 73% of the days in peak periods; in off-
peak periods the risk premia are negative on 75% of the days. For peak periods the
excess capacity on the previous day, financial risk (spot volatility on the previous
day, spread on the previous and current day), and lagged spot prices are found to be
reflected in the risk premia. In addition, the authors calculate the ex ante risk pre-
87
mia. These risk premia are in their magnitude similar to the ex post risk premia but
very sensitive towards the underlying assumptions regarding the spot price model.
Diko et al. (2006) analyze the period January 2001 to August 2005 regarding the exis-
tence of risk premia in three European electricity markets. They use OTC prices and
exchange (intraday) prices. The investigated markets are Germany, the Netherlands,
and France. The OTC prices are provided by Platts, the leading global provider of
energy and metals information. The OTC data consist of day-ahead prices (not
for weekends) for base, peak, and off-peak periods as defined by the corresponding
exchange; the intraday prices are directly obtained from the exchanges. They find
statistically significant positive risk premia in all markets during peak hours. For the
German market they also find negative risk premia in the off-peak period. Further-
more, the authors show that a time-evolution of the risk premia can be observed.
The evolution is expressed in a clear reduction of the risk premia with progressive
maturity of the markets. When analyzing a potential term-structure of risk pre-
mia the results indicate that, as the time-to-delivery decreases, the long-term and
medium-term risk premia increase. The results concur with the Bessembinder &
Lemmon (2002) model.
Hadsell & Shawky (2007) examine the New York wholesale market over the period
January 2001 to March 2005. When analyzing the hourly risk premia the authors
use data from two delivery zones, the New York City and the Genesee zone. Finding
significant risk premia they show that the magnitudes vary on a daily, weekly, and
monthly basis. Furthermore, the risk premia are found to be different in magnitude
across the two zones. The authors argue that this could be due to the fact that
the NYSIO markets are not fully integrated within the wider financial market. The
introduction of virtual bidding is shown to be associated with lower premia in off-
peak hours and higher premia in peak hours. Thus, the effectiveness of speculators
altering the risk premia seems to be limited. This is contrary to the expectations
at the time of implementation of the virtual bidding mechanisms and also to the
empirical results obtained by Saravia (2003) for the first years of operation.
Borenstein et al. (2008) analyze the trading period prior to the occurrence of a
significant event in the still young history of deregulated electricity markets, namely
the collapse of the Californian electricity wholesale market in 2000. Two years prior
to the collapse the Californian market had the two typical market segments, namely
the day-ahead and intraday market, with trading of contracts for the same time
and location. The authors report large price differences between these two market
segments. In addition, they argue that the price differences were ex ante predictable.
However, they exclude the interpretation of this price differences as risk premia
caused by risk aversion due to (i) the regular change in sign, (ii) the opportunity
88
to diversify the risk, and (iii) the high magnitude. They rather attribute the price
differences to trading inefficiencies and market power.
The short-term risk premia in contracts traded for delivery in New England are
examined by Hadsell (2008). The New England market is operated by the New
England ISO and spans over eight zones which are all analyzed by the author. The
dataset consists of the price for peak hours (7am to 11pm) on non-holiday days
in the period January 2, 2004 to December 31, 2007. The main findings are that
the risk premia are positive and consistent over the years. Furthermore, there is
evidence that the risk premia are higher in June and lower in August and October.
The author also reports that the risk premia are stable in their magnitude over the
analyzed period. Furthermore, there is evidence that the ex ante risk premia are
positive in all months.
Ronn & Wimschulte (2009) conduct a first analysis of the German spot market using
data from August 2002 to September 2007 from the day-ahead and block contract
market. In addition, using data from June 2004 to September 2007, the Austrian
market located at the Energy Exchange Austria (EXAA) is analyzed. The posed
question is whether there is a risk premium in contracts traded at the EXAA with
delivery in Germany. This analysis is justified due to the fact that there is no conges-
tion between the Austrian and the German market. Regarding the German market,
the authors find positive risk premia in the block contract market. In particular the
risk premia in contracts with the longest time-to-delivery are found to be high and
statistically significant. For contracts with delivery on non-working days the results
suggest that the risk premia are on average negligible. In addition, the market price
of risk is computed for both investigated markets.
Viehmann (2009) estimates risk premia in the day-ahead market of the EEX. The
author uses price data from the day-ahead market and from the Austrian electric-
ity exchange EXAA. The EXAA data are used as a proxy for data from the OTC
market, in which trading takes place prior to the EEX but price data are not pub-
licly available. The trading in the OTC market mainly takes place – according to
the author – between 8 am and 12 pm on the day prior to the delivery day. Only
standardized block contracts can be traded; the traded volume seems to have the
same magnitude as in the day-ahead market of the EEX. Thus, the OTC-market
can be interpreted as a forward market in relation to the day-ahead market. As the
auction on the EXAA takes place between 10.12 am and 10.15 am and contracts with
delivery in Germany are traded in this auction the OTC prices and the EXAA prices
coincide due to arbitrage possibilities. Consequently the EXAA prices reflect the
OTC market approximately two hours before the auction on the EEX starts. Taking
this point of view the author finds – covering the sample period October 2005 to
89
September 2008 – hourly risk premia that are significantly different from zero and
both positive and negative. However, the overall mean of the daily risk premium is
statistically not different from zero. The largest positive risk premia are found in
evening peak hours during winter months; they amount to around 10%. The average
risk premium in contracts with delivery in four evening hours is shown to be more
than eight times higher during the winter months (November to February) as during
summer months (March to October). This is the time period with the year’s highest
electricity consumption levels.
Long-term risk premia

Shawky et al. (2003) analyze futures traded in the New York Mercantile Exchange
(NYMEX) with delivery at the California-Oregon Border. Their dataset includes
daily data for month futures which start trading six months before delivery. The
data cover the years 1998 and 1999. Trading in the electricity futures started already
in 1996 but due to the low liquidity, the inexperience of the market participants and
the lack of deregulation in other markets the authors exclude the data from the first
two years of operation. As result the authors find positive risk premia which are
estimated to be 0.1328% per day whereby the relative risk premia are defined as the
quotient of the risk premia and the spot price at delivery. The risk premia seem to
increase with increasing time-to-delivery. When analyzing the trading patterns in
electricity futures the authors show that volume gradually increases over the life of
the contracts, i.e. the short-term futures show the highest traded volume. Regarding
the open interest an increase with increasing life of the contracts is also found.
Beyond the 30-day mark a significant drop in open interest appears, indicating that
the majority of the contracts are closed out.
Data from the Nord Pool are analyzed by Lucia & Torro (2008). Their dataset covers
the period January 1, 1998 to October 28, 2007 and consists of the four closest-to-
delivery week futures. Weekly observations of the futures prices are analyzed. The
authors find significant positive risk premia in the short-term futures. It is shown
that the magnitude and the significance of the risk premia varies over the year. The
risk premia are highest in winter months and zero in summer months. The risk
premia are positive 60.5% of the weeks in the dataset. Furthermore, the authors link
the risk premia to unexpectedly low reservoir levels. The authors also show that a
supply shock – probably caused by a significant reduction in hydropower production
due to abnormally low water inflows – hit the market around the end of year 2002 and
changed the market environment. Evidence is reported that prior to the occurrence
of this event the risk premia were related to the variance and to the skewness of the
future spot prices as predicted by the model of Bessembinder & Lemmon (2002).
90
After the supply shock this relation broke down.

Botterud et al. (2010) report results concerning week futures traded at the Nord
Pool. The authors find statistically significant positive risk premia in futures with a
time-to-delivery up to six weeks covering the sample period 1996 to 2006. For their
analysis they use the closing price on the last day of trading in each week for the week
futures and the average weekly price on the day-ahead market. The authors identify
differences between the supply and demand sides in risk preferences and the ability
to take advantage of short-term variation in prices as potential explanations for the
relationship between spot and futures prices. Furthermore, they link the risk premia
to the physical state of the system, expressed by the hydro inflow, reservoir levels
and demand. In addition, the authors argue that due to the high share of hydro
power with large reservoirs not only the hedging pressure theory but also the theory
of storage are relevant for an analysis of the relationship between spot and futures
prices. The convenience yield in the weekly futures contracts is then estimated. The
result is an average net convenience yield which is negative over all holding periods.
The convenience yield is positive in the first half of the year when reservoir levels
are low and negative in the second half when reservoir levels are normally high. Also
the risk premia exhibit a (less distinct) seasonal pattern.
Marckhoff & Wimschulte (2009) analyze Contracts for Difference (CfDs) traded at
the Nord Pool. CfDs are cash-settled future contracts and allow to hedge against
price differences among different delivery areas. This is of interest in the case of
congestion of electricity transmission lines. As a result different prices evolve in
connected grid zones. CFDs were first traded at the Nord Pool at the end of 2000.
The authors analyze the period 2001 to 2006 and find significant short-term positive
risk premia and negative long-term risk premia. These risk premia exhibit significant
variability in terms of sign and magnitude. When testing the model of Bessembinder
& Lemmon (2002) they find that the implications can be confirmed for the risk
premia in the CfDs.
An analysis of the Iberian power futures market is conducted by Capitan Herraiz &
Rodriguez Monroy (2009). The dataset of their study starts in August 2006 and ends
in July 2008. Month, quarter, and year futures are analyzed, however, no statistically
reliable results can be obtained due to this short time period. Thus, the reported
results need to be seen as preliminary. In addition, the market started operation in
July 2006. Problems such as the inexperience of market participants and the low
liquidity question the results even more. The authors analyze the risk premia from
an ex post perspective and show that the futures price is slightly upward biased,
implying positive risk premia. They estimate the risk premia as a percentage of the
futures price. When testing for potential drivers of the risk premia the authors find
91
little evidence for the implications of the Bessembinder & Lemmon (2002) model. In
summary, the authors conclude that energy markets show a limited level of market
efficiency.
Redl et al. (2009) investigate the relevance of systematic forecast errors in the price
formation process in electricity forward markets. The authors analyze data from the
EEX and the Nord Pool. They find that the prices of year futures at the EEX are
influenced by past spot prices. They term this as an adaptive expectation formation
behavior which results in a biased forecasting power of the futures. From an ex
post perspective they show significant positive risk premia in month futures traded
at the EEX. Parts of the risk premia are explained by risk assessment measures.
However, an unexplained part remains and hence inefficiencies are not ruled out by
the authors.
Furio & Meneu (2010) analyze the Spanish electricity market for long-term risk
premia. They use both the ex ante and the ex post approach. Due to the fact
that the Spanish futures market was launched only in July 2006 the authors use
data from the OTC market. They analyze the first-to-deliver month forward which
is a base load contract with the underlying being the Spanish (day-ahead) spot
market. This contract was traded for the first time in February 2003 for delivery in
March 2003. Price data are obtained by Reuters. Covering a sample period between
February 3, 2003 and August 31, 2008 they find that overall the ex post risk premia
are negative but not statistically significant. It is observed that the risk premia
were negative over the period from November 2004 to February 2006 and positive
thereafter. The authors also show that the sign and magnitude of the ex post risk
premia are dependent on the unexpected variation in demand and on the unexpected
variation in the hydroelectric capacity. In addition, both the ex post and the ex ante
risk premia are negatively related to the variance of spot prices as predicted by the
Bessembinder & Lemmon (2002) model. Furthermore, the authors propose a forecast
model for electricity spot prices at a monthly level.
Kilic & Huisman (2010) analyze whether power production flexibility is a substitute
for the storability of electricity. To answer this question they analyze futures prices
from Belgium, Germany, the Netherlands, and the Scandinavian countries, i.e. the
Nord Pool. As results the authors find that futures prices from markets with flexible
power supply behave according to the expectations theory. The hypothesis underly-
ing their study is contrary as production flexibility is supposed to result in futures
prices that are more in line with the theory of storage. Thus, the above relation
is denied. However, their results confirm the existence of risk premia in various
electricity markets.
Wilkens & Wimschulte (2007) report first results on risk premia in the German
92
futures market. The authors analyze the pricing of futures traded on the EEX
between June 16, 2002 and December 31, 2004. They restrict their study to month
futures with a maturity of up to six months. After estimating ex ante risk premia
they compare their results with ex post risk premia. To estimate the ex ante risk
premia they investigate the performance of one- and two-factor reduced-form models.
The authors find positive risk premia, both from an ex ante and from an ex post
perspective. The risk premia are highly volatile and regularly change in sign.
Bierbrauer et al. (2007) aim to provide an overview on the most promising and
established models in the literature for the forecasting of electricity spot prices.
They test these models on data from the EEX and identify three models which best
fit the data. When using the three models for the forecast of ex ante risk premia
they find positive risk premia in the short- and mid-term and negative risk premia
in the long-term contracts.
Estimating the market price of risk is the aim of Kolos & Ronn (2008). For this
purpose, the authors first estimate the risk premia in energy commodities. Using
a dataset from the EEX covering the period 2002 to 2006, the results are positive
risk premia in the case of electricity. This concurs with the results of Wilkens &
Wimschulte (2007).
Chapter 3
A Project Finance Valuation

Tool
Despite the popularity of project finance in practice, there is still only limited relevant
academic literature available, or rather the published literature focuses mainly on
the qualitative aspects of this financing method. This dissertation aims to contribute
to the gradually evolving quantitative literature. Hence, I develop together with two
other doctorate students a project finance valuation tool which is based on stochastic
cash flow modeling and advanced forecast models.
In this chapter I introduce the newly developed project finance valuation tool. In the
fifth chapter the tool will be used to quantify the impact of model complexity on the
valuation results within a case study; in this chapter I focus on a discussion of the
valuation tool itself. However, I also already examine some specifics of the case study.
In particular the cash flow equation underlying the case study is discussed. At the
beginning, I address the general functionality and computational implementation of
the valuation tool. Then, the separate steps of the valuation process are discussed.
Thereby, I distinguish between three process steps, namely input, computation, and
output. I discuss the valuation results, i.e. the NPV distribution and the cumulative
default probability, and their presentation. Furthermore, I discuss the status quo
of the tool and its present limitations. In addition, I focus on the forecast models
applied within the valuation process. An extended excursus addressing time series
modeling and forecasting is therefore provided at the end of the chapter. The forecast
models implemented in the valuation tool are presented and discussed. I conclude
the chapter with a summary of the capabilities of the valuation tool and an outline
of promising avenues for future research.
93
94
This chapter aims to answer the following three key questions:
• What are the intended capabilities of the project finance valuation tool?
• How does the valuation tool work in detail, which forecast models are used for
the computation, and what are the final computation results?
• What are the present limitations of the developed project finance valuation
tool and how could future research help in resolving these limitations?
3.1 Research Question
The newly developed project finance valuation tool is based on stochastic cash flow
modeling. In order to model the future cash flows, I select the stochastic modeling
approach as a sophisticated modeling method, established in the financial literature.
Results obtained through this approach are able to answer questions beyond the
possibilities of other modeling techniques. One of the main advantages is the possi-
bility to estimate ex ante default probabilities of a project finance investment. The
implemented forecast models are based on recent literature and represent the current
standard models. For example, for the forecast of level data these are autoregressive
moving average (ARMA) models, for volatility forecasts the general autoregressive
conditional heteroskedasticity (GARCH) model, and for correlation forecasts the dy-
namic conditional correlation (DCC) model. The research question underlying the
implementation of all these models is how to program an efficient computer algo-
rithm, which is able to deal with these data-intensive computation tasks.
The motivation behind the development of the project finance valuation tool is
twofold. First, the valuation tool is intended to be a general, sophisticated valu-
ation model based on stochastic cash flow modeling and the most recent forecast
models. Second, the valuation tool is intended to be applicable to specific valuation
tasks and still employable by non-trained users. Thus, the tool is to be used for
both research as well as practical projects. Another aim is to design the tool as
general and variable as possible so as to be open for future extensions. In addition,
the tool has been designed in a modular framework in order to allow future exten-
sions by multiple developers. Regarding the valuation of actual projects, the tool is
practicable for the valuation of power plant ventures financed via project finance.
Furthermore, in order to assure meaningful results, the calculation of the cost of
capital is of crucial importance. To obtain unbiased estimates for the NPV, one needs
to take into account the limited lifetime of the project and the resulting changing
capital structure.
95
Another aspect to be taken into account during the development of the tool is that
of the significant data volumes. Both – the input data necessary for the valua-
tion and the data generated within the valuation process – lead to a data-intensive
computation. In addition, correlation structures and non-analytical distributions are
intended to be implemented. The appropriate handling of the resulting data volumes
is critical for an efficient processing time of the valuation tool.
3.2 Functionality and Implementation
3.2.1 Introductory Remarks
The tool developed within this dissertation is intended to valuate project finance
investments. The tool is basically a stochastic cash flow modeling tool.1 On the
basis of a general cash flow equation the tool combines several separate input and
output factors; forecasts for level data, volatilities, and correlations are derived from
various forecast models. It then simulates a predefined number of future cash flow
paths, which are aggregated to probability distributions of the expected cash flows
at certain future point-in-times. Based on these cash flow distributions the tool com-
putes the probability distribution of the expected NPV. In addition, it computes the
cumulative default probability over the project’s lifespan. Moreover, various cover
ratios and default probabilities are calculated for every future point-in-time of inter-
est as well. The computation is performed with Matlab. However, the user specifies
the input parameters and starts the valuation process in Microsoft Excel. The re-
sults are stored in an Microsoft Excel spreadsheet and in an output file. The output
file is displayed at the end of the valuation process. The output file summarizes the
valuation results as well as the underlying input parameters, factors, and the applied
forecast models.
In order to simplify the terminology, I will abbreviate the valuation tool as PFVT
(Project Finance Valuation Tool).
Starting with the development of a basic cash flow tool2 , the functionality of the
PFVT has been extended over time, mainly because of the participation of two of
the involved doctorate students in a research cluster dealing with energy.3 It was first
extended into a valuation tool for project finance and second into a valuation tool
1
Cf. section 2.1.2.5 for a discussion of stochastic cash flow modeling.
2
The PFVT is part of an extended research project. At the moment three doctorate students
are involved in the tool’s development.
3
Two of the doctorate students involved in the development of the PFVT are members of the
research group “Energy 2030”. Both the research group and the doctorate students are supported
by the International Graduate School of Science and Engineering (IGSSE) at Technische Universität
München (TUM).
96
for power plant investments financed via project finance.4 Despite all extensions, the
aim to preserve high flexibility remained valid. Thus, the PFVT can now be used to
value investments both from the equity and from the debt perspective. In general,
the equity side is of more interest when examining capital investments. However, in
the case of project finance, the debt side – as discussed in the second chapter – also
heavily relies on an appropriate modeling of the expected cash flows. Thus, I aim to
cover both perspectives, designing the PFVT in accordance.
The present chapter is primarily devoted to a technical description of the PFVT, as
the fundamentals and specifics of stochastic cash flow modeling are already discussed
in the second chapter. Basically, the functionality of the PFVT can be described in
three broad process steps. I term these process steps also modules, as each process
step is programmed in a separate source code module. The three modules are the
input module, the computation module, and the output module. The input module
is the interface between the user and the PFVT, where the specification of the input
parameters and of the project specifics takes place. The computation module is the
process step where the forecasts and the Monte Carlo simulations are performed.
Finally, the output module is the part where the obtained results are aggregated and
where an output file is computed.
3.2.2 Functionality of the Tool
The aim of this section is to provide an overview of the broad functionality of the
valuation tool and to show the logic of the valuation process. A detailed discussion
of the respective valuation steps is to be found in section 3.3.5
Figure 3.1 depicts the three broad process steps input, computation, and output.
The single process steps within the valuation process are schematically shown in the
figure as well.
The first step of the valuation process consists of the specification of the input pa-
rameters. These parameters comprise the general simulation procedure, the specifics
of the project, and the applicable forecast models.6 After the completion of the
parameter input the user starts the valuation process. No further interventions by
the user are necessary after this step. In a second step, depending on the selected
forecast models in the first step, historical time series stored in the tool are uploaded.
This step is optional. Together with the input parameters, the time series are then
4
Cf. Weber et al. (2010) and chapter 5 for first valuation results obtained by the PFVT.
5
It is the December 2009 version of the PFVT that is described in this chapter and applied for
the valuation of the power plant venture in the case study in the fifth chapter.
6
Cf. figure 3.1 for a summary of the input parameters. Cf. also the excursus in section 3.5 for
a discussion of the implemented forecast models.
97
Figure 3.1
Schematic Functionality Project Finance Valuation Tool
1 2
Input Parameter Specification Time Series (optional)
INPUT
4 3
Monte Carlo Simulation Forecast Generation
COMPUTATION
5
Cash Flow Equation
6
Probability Distributions
OUTPUT
7
Output File
Source: Own work.
processed from the input to the computation module. In this module, in a third
step, the forecasts are computed based either on the specified input parameters or
the historical time series and the selected forecast model. Afterwards, in a fourth
step, the forecast of the single input parameters are used for a Monte Carlo simu-
lation, i.e. the generation of random numbers. The results of this step are linked
in a fifth step with the cash flow equation underlying the valuation process. After
completion of the simulation procedure the resulting data are transferred from the
computation module to the output module where, in a sixth step, the probability
distributions based on the cash flow paths are constructed. In addition, various ra-
tios and probabilities are computed within this process step. Afterwards, these data
are stored both in an output file and in a Microsoft Excel spreadsheet. In a seventh
and last step the resulting output file summarizing the valuation results and the
input parameters is displayed.
Depending on the selected parameters, in particular on the number of iterations, the
whole valuation process can take from a few seconds up to a few hours.7 After the
completion of the valuation process the output file is automatically displayed.
Two steps in the valuation process – the input parameter specification and the linking
of the input factors cased on the cash flow equation – deserve closer attention. These
7
After the input of the relevant parameters the user sees a time bar with the progress of the
valuation task. This enables the estimation of the remaining processing time.
98
two steps are important as adjustments and manipulations within the valuation
process can solely take place within these steps. Other process steps and associated
modules are general parts of the valuation tool, being used independently of the
specific valuation task.
The input parameter specification is the first process step in which the fundamentals
of a valuation task can be manipulated.8 Based on the specified parameter constel-
lation, different valuation results are obtained, allowing the analysis of the valuation
results depending on the specific input parameter combination. Within the case
study, the sensitivity of the valuation results to changes of certain input parameters
is of particular interest. I distinguish between two dimensions of input parame-
ters when examining the results of the case study, namely parameters regarding the
simulation procedure and parameters regarding the applied forecast models.
The linking of the input factors based on the cash flow equation is the second process
step where manipulations are possible. In this step the input factors of a specific
valuation task are combined based on a specified algebraic relation, i.e. the cash flow
equation. A manipulation of the cash flow equation can have a significant impact on
the valuation results. At the moment, the cash flow equation is implemented in a
general form within the valuation tool. The user cannot perform any manipulations
unless the source code has been modified. However, it can, for example, be advan-
tageous to implement a growth factor in one of the input-output-relations as this
mirrors the reality. Some of the desired specifications can be implemented indirectly,
by re-specifying the input parameters. However, a manipulation of the cash flow
equation is a more general and lasting manipulation. Thus, the implementation of a
possibility to manipulate the cash flow equation at the input parameter level is one
of the many potential future extensions of the valuation tool.
3.2.3 Implementation
Li (2000) correctly states that “the concept of simulation is relatively simple, but
writing the computer code to simulate the data and interpreting results are difficult”.9
The implementation of a valuation tool based on stochastic cash flow modeling be-
comes in particular difficult due to the fact that common computer applications as
Microsoft Excel are not practicable for broader stochastic tools. Advanced program-
ming languages need to be used for applications of this kind.
The PFVT is programmed in Matlab, a powerful programming language for numeri-
cal applications. Matlab stands for matrix laboratory and is a numerical computing
8
Default parameters for a particular valuation task can be stored in a separate Microsoft Excel
spreadsheet.
9
Li (2000), p. 87.
99
environment often used in the context of simulation analysis.

The source code of the PFVT consists of separate modules in order to maximize the
flexibility of the tool for future extensions. This flexibility also allows the simultane-
ous further development of the tool in several directions while securing the stability
of the tool.
One of the aims when developing the PFVT was easy handling. This is why the
tool is designed in a way that the user does not need to use Matlab; he or she
can instead calibrate the PFVT via an easy-to-use Microsoft Excel spreadsheet. On
this spreadsheet, the user is able to define all input parameters and then to start
the valuation process. However, the Microsoft Excel spreadsheet merely serves as a
graphical front-end. The whole computation process is performed in Matlab.
3.3 The Valuation Tool
I address the specifics of the valuation process in chronological order. First, I discuss
the input module, afterwards the computation module and at the end of this section
the output module.
3.3.1 Input
3.3.1.1 Input Parameters
Within the input module the user has first to specify the input parameters regarding
the general simulation procedure and second the input parameters and factors of the
specific project.
The necessary simulation parameters mainly concern the Monte Carlo simulation
and the output of the simulation results. The number of iterations and the fre-
quency of the cash flow calculation are of critical importance. Additionally, the
desired frequency of the cash flow distributions reported and the time resolution
which determines how often the cash flow is analyzed need to be specified. The du-
ration of the simulation procedure significantly depends on these input parameters,
in particular the number of iterations has a significant impact on the computation
time.10
The specific project input parameters and factors consist of measures concerning a)
the general project setup, b) the relevant input and output factors, c) the relation
between the input and output factors, and d) the desired forecast models regarding
10
Cf. section 5.3.1 for a discussion of the computation time in dependence of the number of
iterations.
100
the single input factors.

Regarding the specific project parameters the user, in a first step, has to define the
general project setup which includes, among others, the lifespan of the project, the
financing structure of the project, and the depreciation period. In a second and third
step, the input and output factors and the relation between them are specified. In
a fourth step, the factors related to the forecast of the parameters are defined. I
have implemented several forecast models which are shortly introduced in section
3.3.2.2 and then discussed in more detail in section 3.5. Historical time series of
interest rates, exchange rates and other assets are included in the PFVT. These
time series can be used for the estimation of the required parameters for the forecast
models. However, beneath input and output factors with historical time series, the
PFVT allows the manual input of the forecast-relevant parameters as well. The
user may also choose between different distributions, e.g. rectangular or triangular
distributions for the parameter forecasts. This enables the consideration of non-
Gaussian distributed factors being normally one of the main drawbacks of simple
simulation procedures.11
Regarding the number of input and output factors and other relevant factors, the
programming of the PFVT is rather flexible. After a specification of the required
number of factors, the PFVT dynamically adopts to the necessary inputs.
Table 3.1 contains a summary of the parameters concerning the general project setup,
the simulation procedure and the project’s specifics which might be defined by the
user for a specific valuation task.
3.3.1.2 Input Data
Historical data for various interest rates, stock index, and commodities are stored
within the PFVT. The data are obtained by Bloomberg and updated when neces-
sary.12 An extension of the available data is possible when necessary as well.
When the user chooses to forecast a certain parameter based on the historical time
series, the stored data are sent from the storage module to the forecast module.
In the latter module the parameters of the time series, i.e. the mean and standard
deviation during a certain time period in the past, are estimated.13 These parameters
are then used for the forecast process.
11
See for example Nawrocki (2001).
12
The update of the data takes place manually.
13
At the moment, the estimation procedure extends over the whole length of the available his-
torical time series.
101
Table 3.1
Summary Input Parameters
Input Parameters regarding …
… Project Setup … Project Modeling … Simulation Procedure
Initial Investment Interest Euro Life-span Simulation

Interest Dollar Time resolution
Ratio Debt …
Ratio Debt Euro # Iterations
Ratio Debt Dollar Exchange Rate 1
Ratio Debt 3. Currency … Cash flow frequency
Amortization Time Output 1 Price

Tax Rate Input 1 Price
Ratio Input 1 to Output 1
Capacity ….
Load Factor
Availability Factor Fixed Cost
# Outputs
# Inputs
# Fixed Costs
Variable Costs (% Revenue)

Variable Costs per Unit
Source: Own work.
3.3.2 Computation
3.3.2.1 Cash Flow Equation
The PFVT relies on the direct valuation method, implying that no accounting issues
are taken into consideration. The computation is based on the specified input factors
and the underlying cash flow equation.
The cash flow equation is the starting point of an appropriate valuation process. The
equation not only links the specified input and output factors but also introduces
the project’s cost and financing structure into the valuation process.
In general, I distinguish between six constituent parts of a cash flow equation:
1. Revenue,
2. Input factor costs,
3. Variable costs,
4. Fixed costs,
5. Cost of debt, and
6. Tax.
102
Figure 3.2
Underlying Cash Flow Equation
Revenue
Cash Flow =
x
Generation Price El.
Input factor costs
-
Generation x Price Coal x Coal-Factor - Generation x Price CO2 x CO2-Factor
Variable costs
-
Revenue x Percentage + Generation x Constant
Fixed costs
-
Constant
Cost of debt
-
Principal x Interest
Tax
-
Profit x (1-Tax Rate)
Generation =
Capacity x Load Factor x 8760
Source: Own work.
I take the cash flow equation underlying the case study in the fifth chapter to discuss
the single parts of a general cash flow equation. The project of interest is a coal power
plant. Figure 3.2 schematically depicts the cash flow equation. The six general parts
of a cash flow equation defined above are marked with dashed boxes in the figure.
The first part of the cash flow equation, the revenue, consists – in the case of the coal
power plant – of one output factor, namely electricity. The revenue is calculated as
the product of the electricity price and the generated electricity in the period. Both
the electricity price and the generated output are normally modeled as stochastic
factors. In the case study the electricity price is modeled based on the historical
time series and on an appropriate forecast model. The output is calculated as the
product of capacity, the load factor, and the number of hours in the period. The
load factor is a stochastic factor.14
The input factor cost, the second part of the general cash flow equation, consists of
two input factors, namely coal and emission rights. The factor cost is calculated as
the product of the electricity output and the factor cost. The defined input-output-
relation is of critical importance. In the case study, both the coal and emission cost
are modeled based on historical data and on an appropriate forecast model.
14
Cf. section 5.2.1 for a more detailed explanation of the single factors.
103
The third part, the fixed costs, consists of a fixed amount which is defined at the
beginning of the valuation process and then used over the whole lifespan of the
project. These costs are seen as non-stochastic. However, a growth factor (also
negative) or a stochastic evolution can be easily introduced.
The variable cost as the fourth part of the general equation is modeled as a percentage
of the period’s revenue in the period. Normally the percentage remains constant
over the whole lifetime of the project. In addition, a term adding a fixed amount per
generated unit of electricity is implemented.
Of particular importance is the modeling of the cost of debt, the fifth part. In the
case study I assume annuity debt, i.e. the debt is paid back in equal amounts over
the lifespan of the debt contract. In addition, I assume that debt in two currencies is
provided to the project, namely in Euro and in U.S. Dollar. The resulting U.S. Dollar
payments are then variable as they are calculated in each period as the product of a
fixed amount and the current exchange rate.15
The sixth part, the tax, is calculated based on a defined tax rate as the product of
the earnings before tax and the tax rate.
A generalization of the cash flow equation can be achieved through the definition of
further input parameters. Given the possibility to define the number of input and
output factors variable, a second output factor, for example, can be easily added. At
the moment, the design of the tool allows the implementation of five input and five
output factors. However, the complexity and the computation time of the valuation
process grows with the number of input parameters.
3.3.2.2 Input Parameter Forecast
The future development of the factors influencing a project’s cash flow is of huge
importance. Since this development is ex ante unknown, different kinds of forecast
models need to be applied to obtain estimates for the future values. However, not
only the absolute future values of the influencing factors are of interest. To perform
a consistent Monte Carlo simulation, the future variance and correlation structures
between the single input factors have to be estimated as well. The specific models
implemented within the PFVT are discussed in section 3.5, in an excursus on time
series modeling and forecasting. Thus, I only list the implemented forecast models
here.
The following forecast models are implemented within the PFVT:
For mean values:
15
It is intended to implement more flexible debt contract features in the recent future.
104
1. Random walk,
2. ARMA, and
3. Mean reversion.
For volatilities:
1. Historical volatility,
2. GARCH,
3. Glosten, Jagannathan and Runkle – Exponential General Autoregressive Con-

ditional Heteroskedasticity (GJR-GARCH), and
4. Exponential General Autoregressive Conditional Heteroskedasticity (EGARCH).
For correlations:
1. Historical correlation, and
2. DCC.
When a certain forecast model is applied, two possibilities for parameter specifi-
cation exist. First, the parameters are manually specified by the user. Second, the
parameters are estimated based on the historical time series stored within the PFVT.
When the parameters are manually specified the user in general needs to specify the
mean value and the volatility of the relevant factor. In addition, the distribution may
be specified as well, in particular when non-analytical distributions as the triangular
distribution are preferred. Furthermore, the correlations of the relevant factor with
the other factors of interest need to be specified.
When the parameters are estimated based on the historical time series the user
has to specify a certain forecast model. Afterwards, the necessary parameters are
automatically estimated.
3.3.2.3 Monte Carlo Simulation
The Monte Carlo module is the part of the PFVT where in a first step the generation
of random numbers and in a second step the simulation of the cash flow paths takes
place. The stochastic input factors (with the probability distributions generated
within the forecast module) and the non-stochastic factors are combined in this
module to single cash flow paths. The computation time is heavily dependent on the
defined lifespan of the project, the frequency of the cash flow distributions and the
105
number of iterations. At the end, the results are stored in a data matrix and sent to
the output module.
The functionality of the Monte Carlo module consists in a first step in the determina-
tion of a point estimate for each stochastic input factor for every future point-in-time
of interest. Therefore, a random value form the probability distribution of the specific
input factors is determined, e.g. drawn. The combination of these single point-in-
time point estimates based on the cash flow equation results in a single cash flow
estimate. The repeating of this procedure for all selected point-in-times results in a
cash flow path. These cash flow paths are then combined to probability distributions
of the cash flows in the computation module which will be explained later.
The core of the Monte Carlo module is the random number generator. This gen-
erator is critical as it is highly important that the generated random numbers are
independent. I use the random number generators implemented in Matlab. For uni-
form distributions this is a generator developed by Marsaglia (Marsaglia & Zaman
(1991)). For normal distributions, it is a generator developed by Marsaglia as well
(Marsaglia & Tsang (1984)).
3.3.2.4 NPV Calculation
As already discussed in the second chapter, the calculation of the NPV is not a trivial
task due to the non-stability of the relevant discount rate. Usually, the leverage of
a project finance investment changes over time and hence the risk and the cost of
capital of the project change at the same time. Also the determination of the leverage
is not straightforward as it is not clear whether to use market or book values.16
The first problem, the non-stability of the discount rate, is faced within the PFVT
by the recalculation of the discount rate in every point-in-time when a cash flow
probability distribution is calculated. Based on the actual capital structure the cost
of equity are estimated and the WACC then newly calculated. It is to expect that
the WACC decreases with increasing time.
The second problem is solved by applying the quasi-market valuation which is dis-
cussed in section 2.1.2.2.
The NPV calculation routine applies the estimated WACC to every cash flow path
to calculate a point estimate of the NPV. The combination of the point estimates
results in a probability distribution of the NPV.
16
Cf. Esty (1999) for a discussion of these problems.
106
3.3.3 Output
3.3.3.1 Output File
The output module consists of two parts. The first part is responsible for the calcu-
lation of the cover ratios including the default probabilities. The second part serves
the compilation of the output file.
The output file contains three sections:
1. Profitability measures,
2. Cash flow distributions, and
3. Input parameters.
The average volume of the output file is approximately 100 pages. The size of
the output file mainly depends on the selected display resolution; one cash flow
distribution fills one page.
The reported measures in the first and in the second section of the output file are
discussed in the next section. The third section contains a summary of the input pa-
rameter values, as defined by the user or estimated based on the historical time series,
and the underlying forecasts of the individual factors over the valuation horizon.
The summary of the input parameters contains a) an overview of the parameters
concerning the simulation procedure, b) the models and a graphical depiction of all
input parameters defined as stochastic, and c) a graphical depiction of the remaining
debt capital.
In addition, the simulated cash flow paths and the calculated NPV values are stored
in a Microsoft Excel spreadsheet.
3.3.3.2 Output Data
The first section of the output file, the part regarding the profitability measures,
contains (1) the probability distribution of the project’s NPV and (2) the estimated
cumulated default probability.
The second section, the part regarding the cash flow distributions, consists of one
page for each point-in-time for which a probability distribution of the cash flow was
requested. The output page contains:
1. the probability distribution of the free cash flow to equity (FCFE)17 ,

17
The FCFE is the default setting in the PFVT. However, also the FCFF can be estimated and
displayed.
107
2. the default probability in the certain period,
3. the cash-flow-at-risk at three confidence levels,
4. six coverage ratios,
5. two profitability ratios, and
6. two leverage ratios.
In addition, the expected value of the FCFE is quantified and the median value and
the standard deviation are reported as well.
The probability distribution of the FCFE is displayed as a histogram. The x-axis
and y-axis of the histogram are variable to achieve the best possible depiction. The
default probability for the current period and the cumulative default probability are
displayed.
The chosen confidence levels for the cash-flow-at-risk are 1%, 5%, and 10%.
The calculated coverage ratios are (1) EBIT to interest, (2) EBITDA to interest, (3)
DSCR, (4) Debt payback period18 , (5) cash flow from operations (CFO) to Total
Debt, and (6) FCFF to Total Debt. The coverage ratios are discussed in section
2.1.3.2.
The two reported profitability ratios are the gross profit margin, defined as the
quotient of revenue minus variable costs and revenue, and the pretax return on
capital, defined as the quotient of net income and the book value of equity.
As leverage ratios, the long-term debt to capitalization and the total debt to capi-
talization ratio are calculated. The calculation of the ratios is based on book values.
One page of the output file compiled within the case study in the fifth chapter is
displayed in figure 3.3. The number of iterations for the displayed simulation was
500,000, showing the probability distribution of the cash flow after one year.
3.4 Status Quo and Limitations of the Valuation Tool
3.4.1 Status Quo
The status quo of the valuation tool is represented by a stand-alone, computer-based

tool that is specified for the valuation of power plant ventures financed via project
finance.
18
The debt payback period is defined as the quotient of total debt and the FCFE.
108
Figure 3.3
Output File – Cash Flow Distribution with Measures
Source: Own work.

109
The main quantitative results of the valuation tool applied for the valuation of a
project finance investment are:
• probability distributions of the expected future cash flows at certain points-in-

time,
• a probability distribution of the expected NPV,
• cover ratios and default probabilities at certain points-in-time, and
• the cumulative default probability of the project over the entire lifespan.
Within the valuation process, the valuation tool (i) uses various advanced forecast
models for the forecast of level data, volatilities, and correlations and (ii) considers
correlations between all input parameters. Performed robustness checks and the
results of the case study in the fifth section suggest the tool has been well specified.
The implementation of the valuation tool is made as simple as possible, in order
to enable less experienced users to apply it. The relevant parameters are typed in
a Microsoft Excel spreadsheet. After the start of the valuation process, the whole
process is executed automatically, generating an output file with all results and a
summary of the input parameters.
3.4.2 Limitations
The present limitations of the PFVT are mainly related to three aspects. First, the
PFVT in its current stage is developed and calibrated for the valuation of power
plant ventures. Second, the tool assumes the same discount rate over all cash flow
paths. Third, hedging is not yet implemented.
The first point relates to the question whether the obtained results are of general
validity. In the case study in the fifth chapter, the focus is on the results’ relative
robustness and not on the absolute valuation results. Hence, it would be of interest to
apply the valuation tool on projects in other sectors in order to verify the accuracy
of the results. Projects in the infrastructure or the telecommunication sector, for
example, would be suitable. However, public data availability unfortunately is low.
It is intended to obtain data from practitioners active in this field in the near future
to extend the range of applications.
The application of the same discount rate, i.e. the use of the same operational risk
or unlevered cost of equity, on all cash flow paths could lead to a biased estimate
of the NPV. The fact that normally a higher volatility is linked to higher risk and
hence higher cost of capital has to be taken into account. The theory of risk-neutral
110
valuation promises a solution. According to this approach, the cost of capital must
be calculated separately for every cash flow path to ensure the theoretical accuracy
of the valuation. The implementation of an advanced cost of capital procedure is
one of the next intended improvements of the PFVT.
The implementation of hedging would allow for specifying hedging strategies within
the individual valuation tasks. As hedging is a common phenomenon observed in
project finance investments, as is discussed in section 2.1.1.5, the result would be a
more realistic specification of a valuation task. Hedging strategies regarding the costs
of input and output factors, currency rates, and interest rates could be introduced.
In the case of a power plant venture, the hedging of the only output, the electricity,
based on the results in chapter 4 regarding the risk premia, i.e. the bias between
futures prices and the expected spot prices, is of interest. The main advantage of
hedging is reducing future cash flow variability. As an implementation of hedging
requires a significant reprogramming of the PFVT, hopefully the next generation of
doctorate students will solve this drawback.
3.5 Excursus: Time Series Modeling and Forecasting
3.5.1 Introduction
The results of a valuation process based on stochastic cash flow modeling rely heav-
ily on the appropriate specification of the input factors and, in particular, on the
applied forecast models. The forecast models implemented within the PFVT are
based on time series modeling. This excursus serves as a short introduction to this
topic and aims to clarify the rationale, the advantages, and the potential drawbacks
of the implemented forecast models. However, this excursus does not provide a com-
prehensive overview. I solely address the forecast models implemented within the
PFVT and I refer the interested reader to continuative literature regarding further
models.19 Furthermore, I try to maintain a low level of mathematical complexity in
this section and rather focus on the intuition behind the forecast models of interest.
A time series is generally defined as a sequence of observations (data points) chrono-
logically ordered according to their time of occurrence. The time intervals between
the single observations are typically uniform. Time series modeling refers to the anal-
ysis of historical time series with the aim of identifying their characteristics. These
characteristics can then be used for the forecasting of future values. Often, the same
19
Cf. for example Geman (2005) and Weron (2006) for an overview on forecast models regarding
commodity and electricity markets. A more detailed discussion on time series modeling and fore-
casting can be found, for example, in Brooks (2002) on an introductory level and in Pena et al.
(2001) and Tsay (2002) in a more sophisticated context.
111
mathematical model is used for modeling and forecasting purposes. Therefore, in

a first step, an appropriate model that describes the behavior of a time series is
identified, and then, in a second step, it is used for the forecast of future time series
values.
In the following I discuss forecast models for level data, volatilities, and correlations.
The forecast of all these three time series characteristics is necessary in a typical
valuation process. For the forecast of level data three models can be used within the
PFVT: (1) the random walk (with and without drift), (2) ARMA, and (3) the mean
reversion model. For the volatility forecast the implemented models are: (1) histori-
cal volatility, (2) the general autoregressive conditional heteroskedasticity (GARCH)
model, (3) the GJR-GARCH model, and (4) the EGARCH model. For the correla-
tion forecast, there are two models implemented: (1) historical correlation and (2)
the dynamic conditional correlation (DCC) model.
In the next sections, following symbols are used if not differently stated: the value
of a time series at time t is noted as yt , the value of the same time series at time t-1
as yt−1 . The first difference of a time series at time t is rt . The volatility of a time
series at time t is noted as σt .
3.5.2 Forecast Models Level Data
I introduce the random walk model with and without trend, the ARMA model class,
and the mean reversion model as forecast models for level data.
Random walk model

The random walk model is part of an important class of stochastic processes20 mainly
used for the modeling of non-deterministic time series. The Brownian motion, also
known in the mathematical literature as Wiener process, is the continuous-time
version of the random walk.
A simple random walk is defined as
yt = yt−1 + µ + ut (3.1)
with µ as a constant drift component parameter and ut a white noise process.

The white noise process is a standard stochastic process often applied to model
random shocks in times series. White noise is defined as a stationary21 discrete
20
A stochastic process can broadly be defined as a time series variable that is evolving in a random
way.
21
A time series that exhibits stationarity is characterized by the properties that the mean, vari-
ance, and autocorrelation structure are constant.
112
stochastic process, i.e. a process of uncorrelated random variables with an expected

value of zero and with a constant volatility. A white noise process has no discernible
structure. Normally, it is assumed that the white noise exhibits a normal distribution.
The general continuous notation of a change in the time series variable y as a differ-
ential equation of the generalized Wiener Process is
dy = adt + bdz (3.2)
with a as a trend parameter and b as a volatility parameter.

Formula (4.2) states that the value yt of a time series at time t equals the value one
time period before plus a random change (random walk without drift). In the case
that µ is unequal to zero a deterministic change is in addition applied (random walk
with drift).
An important characteristic of the random walk is the fact that it follows the Markov
property. This property implies that the future values do not depend on the current
or past values. Thus, the expected value of a random walk without drift is its actual
value. For a random walk with drift the expected value is the actual value plus the
drift component times the incremental time step.
The use of the random walk is in particular popular in finance due to the market
efficiency theorem and its implications. According to this theorem the evaluation
of market prices, i.e. stock prices and other financial asset prices, is best described
by the random walk with or without drift. Also the fundamental models in option
pricing are based on the random walk.22
Autoregressive moving average (ARMA)

The ARMA model class consists of linear models for stationary and discrete stochas-
tic time series. I address the general ARMA model in three steps. First, I discuss the
first part of the general model class, the autoregressive (AR) models. Second, the
moving average (MA) model class is introduced. Finally, the combination of these
two model classes to the ARMA model class is addressed.
AR models
The basic assumption underlying AR models is the existence of a temporal relation
between the values of a time series. In this case, the actual value of a time series can
be to a certain degree explained by a linear function of its past values. The number
22
Cf. Hull (2008) for a discussion of the application of random walks in the modeling of financial
data.
113
of past values, denoted as p, determines the order of the AR model.

An AR model of order p, abbreviated as AR (p), is noted as
p
X
yt = α + ai yt−i + ut (3.3)
i=1
where α and the ai are constant parameters and ut is the white noise process discussed
above.
The simplest AR model is the AR (1) model, defined as
yt = α + a1 yt−1 + ut . (3.4)
The AR (1) process is a discrete version of the continuous mean reversion process
discussed below.
The current value of a time series modeled with an AR model is the sum of the past
weighted values of the time series, an error term, and a constant.
MA models
MA models use the past noise terms of a time series to model the future values.
An MA model of order q, abbreviated as MA (q), is defined as
q
X
yt = β + bj ut−j + ut (3.5)
j=1
with β and bj as constant parameters.

The simplest MA model, the MA (1) model is noted as
yt = β + b1 ut−1 + ut . (3.6)
The MA (1) model is the random walk discussed above.

The current value of a time series modeled with the MA model class is the weighted
average of the q processing noise terms and an error term.
ARMA models
The combination of the two above discussed model classes, namely the AR and MA
models, results in the ARMA model class. ARMA models are linear models for
114
stationary time series. The general ARMA (p, q) is defined as

p
X q
X
yt = χ + ai yt−i + bj ut−j + ut (3.7)
i=1 j=1
with χ as a constant.
The simplest ARMA model is the ARMA (1, 1) which can be noted as
yt = χ + a1 yt−1 + b1 ut−1 + ut . (3.8)
The ARMA models were introduced by Box & Jenkins (1970) in econometrics. To-
day, the use of the ARMA model mainly focuses on the ARMA (1, 1) model as it
yields good forecast results with manageable modeling complexity.23
Mean reversion model

Mean reversion is a term for a stochastic process used in the financial literature
which is known in the mathematical literature as the Ornstein-Uhlenbeck process.
The process is a continuous stochastic process defined as
dyt = θ(µ − yt )dt + σdWt (3.9)
where θ, µ, and σ are (constant) parameters and Wt being the Wiener process.
The main characteristic of the mean reversion model is already included in its name:
Time series modeled with this approach tend to return to the long-term mean value
represented by µ in the above equation. Θ is a rate parameter determining on which
time scale the times series variable is converging towards µ.
Mean reversion is an effect often observed in economic time series. It describes
the tendency of certain time series to return to their long-term average. This is of
particular interest for commodities, as parts of the academic literature assume that
commodity prices mean-revert to long-run equilibrium prices.24
Vasicek (1977) applied the Ornstein-Uhlenbeck process for financial data, introducing
a one-factor model for the evaluation of interest rates to the financial literature.
Since the work of Vasicek’s many applications of the mean reversion characteristic
for financial data have been found. Thus, this model belongs to the standard time
series models applied in finance.
23
Cf. for example Neusser (2009) for a discussion of the parameter estimation for a general
ARMA (p, q).
24
Cf. for example Schwartz (1997) for a discussion of this assumption.
115
3.5.3 Forecast Models Volatility
The above discussed models regarding the mean or level assume homoscedasticity; in
other words, a constant variance and covariance function is assumed.25 However, in
reality this assumption is rarely justified.26 In this section, I introduce models that
are able to deal with time-varying volatility. I discuss the historical volatility ap-
proach, the basic GARCH model, and advanced models related to the basic GARCH
model, namely the GJR-GARCH and the EGARCH.
Historical volatility
The historical volatility approach consists in the forecast of the future volatility
based on the past values of a time series variable. Another term for this approach is
realized volatility. The historical volatility is calculated as the standard deviation of
the past returns.
To calculate the historical volatility, three parameters must be specified:
1. data frequency,
2. time interval, and
3. time period.
Data frequency27 refers to the time interval between the single observations of a
time series. In general, time series have a daily, weekly, or monthly frequency.28
The choice of the data frequency has a significant impact as the estimation results
can be very sensitive to this parameter.29 The time interval specifies the number
of observations that are included in the calculation. This parameter is in particular
critical when the chosen time interval is either too short or too long. In the former
case the selected time period can be randomly uncharacteristic for the modeled time
series. In the latter case the too long time interval can lead to a negligence of recent
changes in the time series characteristics as they are averaged out with the past
data. The time period is finally the period over which the standard deviation is
calculated. This parameter is critical due to the fact that financial time series data
are not stable, i.e. the properties of a times series – inter alia the volatility – change
25
Cf. Weron (2006), p. 113.
26
Cf. Mandelbrot (1963) and Fama (1965) for first empirical results questioning the constant
volatility assumption for financial data.
27
Depending on the data frequency the volatility must be annualized. Since the variance is linear
in time the volatility evolves with the square root of time.
28
Stock prices are, for example, typically available with a daily frequency, whereas macroeconomic
data are mostly available with a monthly frequency.
29
Electricity prices are a good example due to electricity’s diverse seasonality.
116
over time. Similar to the case of the too short time interval using a time period
that is untypical for the time series of interest can also result in misleading volatility
estimation.
When the historical volatility approach is in practice applied the rolling window
method is often used. A fixed time interval within this method is specified for the
estimation of the volatility; after the inclusion of a new observation the oldest data
point is excluded. A time interval between 90 and 180 days is usually selected, based
on daily data. The time period ends with the last available observation.
Advantages of the historical volatility approach are its easy application and the in-
tuitive understandability.
GARCH
The general autoregressive conditional heteroskedasticity (GARCH) model was in-
troduced by Tim Bollerslev in 1986. Nowadays it can be seen as the standard model
for volatility modeling and forecasting, both in the academic world and in real life
applications. The work of Bollerslev (1986) is based on Engle (1982) who developed
the autoregressive conditional heteroskedasticity (ARCH) model.30 In the developed
valuation tool both the standard GARCH model and the extensions GJR-GARCH
as well as E-GARCH are implemented. I first discuss the ARCH model before I ex-
amine the GARCH model.31 The derivatives are discussed thereafter. An excellent
survey on this topic can be found in Bollerslev et al. (1992).
ARCH models
Robert Engle introduced the ARCH process in 1982 as an approach to model the
variance of a time series (Engle (1982)).32
An ARCH process of order one, noted as ARCH (1), is defined as
rt = σt Xt (3.10)
with
σt2 = ω0 + α1 rt−1
2
(3.11)
30
Tim Bollerslev was a doctorate student of Robert Engle.
31
Cf. Brooks (2002) for a discussion of the parameter estimation procedures for ARCH and
GARCH models.
32
In 2003 Robert Engle received (together with Clive Granger) the Nobel Memorial Prize in Eco-
nomic Sciences “for methods of analyzing economic time series with time-varying volatility (ARCH)”;
Diebold (2004), p. 165.
117
as a process for the conditional variance. σt , ω0 , and α1 are positive and constant
parameters; Xt is an independent and identically-distributed random variable, i.e.
white noise. It is normally assumed that Xt is standard normally distributed.
The general ARCH (p) process is defined as
p
X
σt2 = ω0 + α1 rt−1
2 2
+ ... + αp rt−p = ω0 + 2
αi rt−i . (3.12)
i=1
The first parameter in equation (3.12) is the weighted unconditional variance

p
X
ω0 = (1 − αt )σ 2 > 0. (3.13)
t=1
The second parameter (class) in equation (3.12), the αt ’s, are constant and equal or
larger than zero.
The conditional variance in the general model is a weighted sum of squared observa-
tions of the time series. It is ensured that the most recent observations are weighted
more if the parameters in the sum are chosen in a certain way.
Two advantages of the ARCH model are: older information of a time series does
not get lost and more present data is weighted more. Further advantages are the
intuitive understanding and the relatively simple implementation. However, there
are limitations of the ARCH models: the number of past values, used to estimate the
variance, tends to be very high, resulting in a large and difficult to handle model. In
addition, there is no clear best approach to determine the value of q. Furthermore,
the non-negativity constraints might be violated.33
GARCH models
Tim Bollerslev introduced the GARCH model in 1986 (Bollerslev (1986)). As already
implied by the name the GARCH model is a generalization of the ARCH model. The
generalization is that the conditional variance depends on its own history.
Similar to above the GARCH (1, 1) model is defined as
σt2 = ω0 + α1 rt−1
2 2
+ β1 σt−1 . (3.14)
33
Cf. Brooks (2002), p. 452.
118
The general GARCH (p, q) model is defined as
σt2 = ω0 + α1 rt−1
2 2
+ α2 rt−2 2
+ ... + αp rt−p 2
+ β1 σt−1 2
+ β2 σt−2 2
+ ... + βq σt−q
p q
X
2
X
2
(3.15)
= ω0 + αi rt−i + βj σt−j .
i=1 j=1
The conditional volatility is modeled within the GARCH model as a linear combi-
nation of a constant, the sum of weighted squared past errors, and weighted past
values of its own.
The generalization of the ARCH model is that in addition to the past values the
estimated variance is included.
Forecast results obtained by the GARCH model are characterized by either a mono-
tone convergence or stability. In the long term the volatility forecast converges to
the conditional volatility.
The main drawback of the GARCH model is that it can not capture the asymmetric
impact of negative and positive returns on the volatility. GARCH models assume
that both positive and negative returns have the same effect on the volatility. How-
ever, empirical studies show that negative returns tend to indicate higher volatilities
than positive. The economic rationale behind this observation is probably that neg-
ative news have a stronger impact on the market than positive news.34
The discussed drawback can be overcome by one of the two models introduced in
the following.
EGARCH
The EGARCH model was proposed by Nelson (1991). The motivation was an over-
coming of the main weakness of the GARCH model, namely its failure to capture
asymmetric volatility effects.
The general EGARCH (p, q) process is defined as35
r
rt−1 |rt−1 | 2
ln(σt2 ) =ω+ 2
βln(σt−1 ) + γq + α[ q − ]. (3.16)
2
σt−1 2
σt−1 π
The reader is referred to the original work of Nelson (1991) for a discussion of the
individual parameters.
Besides the model’s advantage to capture the asymmetries, the fact that the con-
34
In the case of stock returns, this effect is called leverage effect and was introduced by Black
(1976b) to the financial literature.
35
Cf. Brooks (2002), p. 470.
119
ditional variance is modeled logarithmically causes the improvement that no non-

negativity constraints on the model parameters must be imposed.
GJR-GARCH
Glosten, Jagannathan, and Runkle developed an extension of the GARCH model
which accounts for possible asymmetries (Glosten et al. (1993)). The model is named
after the authors as GJR model or GJR-GARCH model.
Within this model the negative and positive returns are separately estimated to
capture the asymmetric effects. The proposed model contains a dummy variable
which is defined as

1 if rt−1 < 0
It−1 = (3.17)
0 otherwise
The GJR-GARCH model is noted as36
σt2 = ω0 + α1 rt−1
2 2
+ βσt−1 2
+ γrt−1 It−1 . (3.18)
The reader is again referred to the original work of Glosten et al. (1993) for a dis-
cussion of the individual parameters.
The GJR-GARCH model extends the GARCH model at one additional term which
accounts for the asymmetries.
3.5.4 Forecast Models Correlation
Correlation is a measure of linear relationships between two or more time series or

random variables.37 An appropriate estimation of the correlation structure within
the input parameters in the valuation process is of immense importance for the speci-
fication of a realistic valuation task.38 I discuss and implement two common forecast
36
Cf. Brooks (2002), p. 469.
37
To be more precise, it is the correlation coefficient that is a quantitative measure for the linear
relationship. The Pearson correlation coefficient ρX,Y for two random variables X and Y defined as
E[(X − E(X))(Y − E(Y ))]
ρX,Y =
σX σY
is mostly used. σX and σY are the standard deviations of X and Y, respectively. E is the expectation
operator.
38
The PFVT is intended to be mainly applied in the context of energy projects. A good example
for the impact of correlations on the valuation results is a project where both oil as well as gas
are an input or an output factor. In the past, it was common to couple the gas price on the oil
price, i.e. a strong positive correlation between these two time series was apparent. A negligence
or misspecification of the correlation between these two commodities can hence result in valuation
120
models, the historical correlation approach and the DCC model.
Historical correlation
The historical correlation approach is similar to the above discussed historical volatil-
ity approach. Based on two historical time series the correlation coefficient is esti-
mated within this approach. The obtained correlation coefficient is then used as a
forecast for the future correlation.
The correlation coefficient is estimated based on stationary time series. Thus, the
first difference of financial time series is used for the estimation, e.g. the return time
series instead of the stock price time series is relevant. The obtained estimator for
the correlation is a point estimate. All past observations included in the estimation
procedure are equally weighted.
Analog to the historical volatility approach the estimation result is sensitive to the
three above discussed parameters, namely (1) the data frequency, (2) the time in-
terval, and (3) the time period. I refer to the section about the historical volatility
approach for a discussion of the parameters and their impact on the estimation re-
sults. I only emphasize that in the case of correlations the time instability of financial
time series parameters is even more critical. The rolling window method is also often
used in the application of the historical correlation method.39
The use of the historical correlation approach is common in real life applications, in
particular for long-term forecasts. A drawback of the approach is its limitation to
only two time series.
Dynamic Conditional Correlation

The DCC model was introduced in 2001 by Engle and Sheppard.40 The DCC model
is a multivariate GARCH model, i.e. a non-linear combination of univariate GARCH
models. The model aims to estimate the correlation dynamically. The specification
of the model enables the simultaneous estimation of the correlation between arbitrary
many time series.41
The correlation estimation via the DCC model is a two step procedure. In a first
step, the parameters of a univariate GARCH model for every time series included
results that are completely unrealistic due to the coupling.

39
Cf. the previous section for a discussion of the rolling window method.
40
Cf. the working paper of Engle & Sheppard (2001) for an introduction of the DCC. The model
was finally published in the Journal of Business and Economics Statistics in 2002 by Engle; cf.
Engle (2002).
41
The Constant Conditional Correlation (CCC) model proposed by Bollerslev in 1990 can be seen
as the ancestor of the DCC model. It is also based on a similar parameter estimation procedure.
However, the model assumes constant correlation.
121
in the estimation procedure are estimated separately. Standardized errors based on

the estimation results are then calculated. In a second step, the correlations are
estimated based on the standardized errors.
I forgo to provide a mathematical definition of the DCC model as it would go be-
yond the scope of this dissertation. The interested reader is referred to the original
publication of Engle (2002) as well as to the excellent survey of Bauwens et al. (2006).
Similar to the volatility forecasts performed via a GARCH model correlation forecasts
performed by the DCC models tend to be either stable or to converge to the long-
term average of the time series. Thus, in the long term the results are identical to the
results obtained by the historical correlation approach. On average the convergence
takes place between a few weeks and a few months.
Advantages of the DCC model are its appropriateness for financial data as it cap-
tures its characteristics and the ability to model the correlation dynamically. The
correlation is dependent on the standardized returns and on its past. A drawback,
in particular from this dissertation’s point of view, is the fact that in the long term
the results of the DCC converge to the results of the simple historical correlation
method. This poses the question whether the high complexity of the DCC model
justifies its application since the parameter estimation is difficult and data-intensive.
3.6 Concluding Remarks and Future Research
The newly developed project finance valuation tool is introduced in this chapter.
The tool is based on stochastic cash flow modeling and advanced forecast models.
This tool has been developed for the valuation of project finance investments and
will be used for the valuation of a power plant venture in the fifth chapter.
The tool valuates project finance investments from the perspective both of an equity
and a debt provider. The tool uses various forecast models for the separate input
and output factors, i.e. level data, volatilities, and correlations, and combines these
factors on the basis of on a predefined cash flow equation. Then a predefined number
of future cash flow paths is simulated; these paths are aggregated to probability
distributions of the expected cash flows. The probability distribution of the expected
NPV is computed based on these distributions. Besides that, the cumulative default
probability over the project’s lifespan is derived as one of the main results of the
valuation process.
The main results are probability distributions of the expected NPV, expected future
cash flows, and the cumulative default probability over the lifetime of the project.
The tool in its current stage is a valuation tool for single projects. As this is not
122
a limitation for the questions posed in this dissertation, it is of interest to extend

the valuation tool to the application in the context of a portfolio of projects in the
future. In particular in the case of power plant ventures, the examination of whole
power plant fleets is of interest.
The implementation of further forecast models such as futures-based models for
volatility forecasts is also intended. This model class allows the introduction of
recent market expectations in the valuation process. Therefore, the actual market
expectations are taken into account and the valuation results may become more
realistic. In addition, hedging needs to be implemented to obtain even more realistic
results.
Furthermore, an extension of the valuation tool for real options42 seems to be inter-
esting and promising. The NPV method is based, among others, on the assumption
that future actions are already determined at the beginning of a project. Real op-
tions analysis abandons this assumption and allows for future flexibility.43 Thus, a
real option is defined as “the right, but not the obligation, to take an action (e.g.,
deferring, expanding, contracting, or abandoning) at a predetermined cost called the
exercise price, for a predetermined period of time – the life of the option”.44 Al-
ready in 2001, as reported in the survey of Graham & Harvey (2001), 27% of the
participants answered that they (always or mostly) apply real options when valuing
important capital investments. It can be assumed that this number continues to
grow, as the advantage of the real options analysis is apparent from a theoretical
point of view. In particular for energy projects, where future flexibility plays a major
role, real options have the power to significantly increase the insight into the project.
Thus, it is intended to implement real options into the valuation tool as soon as
practicable.
42
Cf. for example Dixit & Pindyck (1994) and Copeland & Antikarov (2001) for profound intro-
ductory literature on real options.
43
Cf. Erner et al. (2003) for a discussion of the advantages of the real options approach compared
with the traditional valuation approach based on the NPV.
44
Copeland & Antikarov (2001), p. 5.
Chapter 4
Price Formation in the German

Electricity Wholesale Market –
An Empirical Analysis
According to the theoretical and empirical literature the risk premia approach seems
to be the most promising theoretical foundation for price formation in electricity
futures markets. As the commodity electricity is non-storable, all electricity con-
tracts are forward contracts. In particular the exchange-traded day-ahead market
contracts, which are commonly termed as spot contracts, are consequently future
contracts with a time-to-delivery of one day.
In this chapter I conduct an empirical analysis of the German electricity wholesale
market. I aim to determine whether there is evidence for the risk premia approach
being an appropriate price formation mechanism. I analyze and report the obtained
results for the spot and futures market separately. Regarding the spot market I an-
alyze all three market segments: the block contract market, the day-ahead market,
and the intraday market. These market segments were or are in operation since
the foundation of the German wholesale market in its current form in 2002. At
the beginning, I discuss the underlying research question, the data, and the applied
methodology. Then, I address the liquidity of the individual market segments. After-
wards, I report descriptive statistics for the spot and the futures market, and analyze
the existence of seasonality in the price time series and the occurrence of negative
prices in the spot market. I conclude the chapter with a summary of the results
regarding the risk premia in spot and futures contracts and an outline of promising
avenues for future research.
123
124
• Is there evidence for the suitability of the risk premia approach as theoretical
price formation mechanism in the German electricity market?
• Specifically, is there empirical evidence for existence of risk premia in the Ger-
man electricity market?
• In the case of empirical evidence for risk premia, what are the properties of
these risk premia and is it possible to identify potential drivers?
Theoretical and empirical literature identify the risk premia approach as the most
promising theoretical foundation of price formation in electricity future markets.1
The following analysis of the German electricity wholesale market is related to that
literature. I focus on the empirical verification of the risk premia approach for the
German market. The research question in this chapter is whether there is evidence
for the suitability of the risk premia approach as a price formation mechanism for the
German electricity market. Thus, the focus of the analysis is on empirical evidence
for risk premia in the exchange-traded electricity contracts.2
Empirical evidence for the adequacy of the risk premia approach is found for other
electricity wholesale markets. Similar findings for the German market would imply
that it exhibits characteristics parallel to these markets. Furthermore, these findings
would imply that the observed electricity prices are biased estimators of the expected
future spot price.
I conduct an in-depth empirical analysis of the German electricity market, analyzing
all market segments3 that are or were active in the last eight years. I divide this
analysis into two parts. The first part deals with the spot market, and the second part
with the futures market.4 I analyze the risk premia from an ex post perspective. An
analysis from this point of view relies on the assumption that the market participants
1
Cf. section 2.3 for a discussion of the theory behind price formation in electricity futures
markets and of the empirical evidence.
2
Cf. Schnorrenberg (2006) for a first analysis of price formation in the German forward market
and for a discussion of other potential price formation mechanisms.
3
The options market is not considered in the following analysis of the German electricity whole-
sale market due to the observed low liquidity. Furthermore, the focus of the analysis is on price
formation in electricity futures markets. The liquidity in the options market has been low since the
beginning and no real improvement is observed over the years. In 2009, for example, trading in the
options market took only place on 30 of 252 trading days.
4
Cf. also Pietz (2009a) for results regarding the German futures market and Pietz (2009b) for
results regarding the German spot market.
125
form their forecasts based on rational expectations.5 This assumption ensures that
the ex post risk premia are on average equal to the ex ante risk premia.6
Regarding the German spot market, the research of this dissertation is related to the
work of Ronn & Wimschulte (2009) and of Viehmann (2009). Ronn & Wimschulte
(2009) conduct to my best knowledge the first empirical analysis of the block contract
market located at the EEX; Viehmann (2009) conducts to my best knowledge the
first empirical analysis of the day-ahead market located at the EEX. In the following,
I confirm the results obtained for the block contract market; I find similar results
for the day-ahead market. However, I take another point of view from Viehmann
(2009) who uses the day-ahead market prices as spot prices in relation to OTC
prices.7 I estimate the risk premia in the day-ahead market contracts in relation to
the intraday market contracts; thus the day-ahead market is regarded as the futures
market. By extending the sample period as well as by including the intraday market
in this analysis I am also able to answer additional questions regarding the German
spot market.
Previous research regarding the German futures market has been conducted by
Wilkens & Wimschulte (2007). The authors use data ranging from 2002 to 2004
and find evidence for positive risk premia in the German futures market. I extend
this research by using a larger dataset, spanning almost eight years. The data are
hence characterized by a higher level of market liquidity and a more mature market
environment, which is expected to lead to more meaningful and robust results.
I contribute with my research in at least a threefold way to the existing literature.
First, to my best knowledge, I am the first to conduct an in-depth analysis of the
German intraday market. By analyzing a sample period covering 30 months, I believe
that first empirical conclusions can be drawn.8 Second, the sample period in which
all three market segments of the German spot market were simultaneously existent
gives me a unique opportunity to investigate the existence of a term structure of
risk premia on a very short time scale. Third, through the analysis of the German
futures market, I am able to contribute to and to extend the empirical literature and
the ongoing discussion on the magnitude and on the sign of potential risk premia in
electricity future contracts and on the evolution of these risk premia over time.
5
Cf. section 4.3 for a critical discussion of this assumption.
6
Cf. section 4.3 for a discussion of the applied methodology.
7
Viehmann (2009) uses price data from the Austrian exchange as proxies for OTC prices regard-
ing electricity delivery in Germany; cf. section 2.3.3.2 for a discussion of his methodology.
8
Trading in the intraday market started in September 2006. However, I exclude the first 16
months of trading due to the low liquidity; cf. section 4.4.1.1 for an analysis of market liquidity.
126
4.2 Data
I use price and volume data from the German electricity wholesale market spanning a
time period between January 2002 and June 2010. Data from the day-ahead market,
the intraday market, the block contract market, and the futures market are available.
All data have been directly obtained from the EEX.9
The day-ahead market data consist of hourly prices and the corresponding traded
volume, covering the period between July 1, 2002, and June 30, 2010. Prices for
the day-ahead market and all other market segments are quoted in Euro/MWh. To
simplify the terminology I will report prices in Euro only. The term hour 1 and hour
1 contract is used for the hour contract with delivery between midnight and 1 am;
the following hour contracts are termed accordingly. The data for the day-ahead
market are available 365 days a year. The daily and monthly Phelix Base and Phelix
Peak10 are already computed by the EEX; they are included in the original dataset.
Data from the block contract market are available between August 1, 2002 and
August 31, 2008. The last day of the dataset is at the same time the closing day of
this market segment. The dataset includes price series of the three block contracts
traded in this market segment (base load, peak load, and weekend base load) and the
traded volume. Each price time series consists of volume-weighted average prices.11
Data from the intraday market are available for the period September 25, 2006 to
June 30, 2010. The first day of the dataset is at the same time the first day of trading
in this market segment. The dataset includes hourly prices and the corresponding
traded volume. Two prices are available for every hour contract in the intraday
market, the average price and the last price. The average price is the average of
all prices from trades which took place in a specific hour contract over its trading
period.12 The last price is the one at which the last trade took place. No information
on the point-in-time of the individual trades is available. I discuss the question which
of the time series should be used for the empirical analysis in section 4.4.2.1.
The futures market data cover the period between July 1, 2002, and June 30, 2010.
Price data for the month, quarter, and year futures are available. In addition, start-
ing March 24, 2010, data for the newly introduced week futures are available, but
not included in the dataset on account of the short sample period. The futures
market data consist of daily prices. In addition to the price data, the open interest
and the traded volume for every future are included in the dataset. The volume
9
Parts of the data are not publicly available. A temporary access to the server of the EEX has
to be purchased to obtain these data.
10
Cf. section 2.2.4.2 for details regarding the calculation of the Phelix.
11
Cf. EEX (2009b), p. 7.
12
Cf. section 2.2.4.3 for a discussion of the trading process in the intraday market.
127
Table 4.1
Summary Data
Summary of the analyzed data. The spot market data consist of price and volume data for three market segments, namely the
day-ahead market, the intraday market, and the block contract market.
Market Segment Data
Day-ahead Market July 1, 2002 - June 30, 2010
Price and Traded Volume Data for every Hour Contract
Data for all Calender Days
Block Contract Market August 1, 2002 - August 31, 2008

SPOT
Daily Price and Traded Volume Data
MARKET
(three block contracts: base load, peak load, weekend base load)
Data for all Exchange Days
Intraday Market September 25, 2006 - June 30, 2010
Price and Traded Volume Data for every Hour Contract
Data for all Calender Days
Futures Market July 1, 2002 - June 30, 2010
FUTURES Daily Price and Traded Volume Data (month, quarter, year futures)
MARKET Daily Open Interest Data
Data for all Exchange Days
Source: Own work.
data are also available for OTC trades which are cleared by the EEX. Only futures
with financial settlement are included in the dataset as the liquidity of futures with
physical settlement is very low.
Table 4.1 summarizes the data and its main characteristics. In section 4.4.1 the
liquidity of the individual market segments is discussed and the choice of the data
used for the empirical analysis explained.
4.3 Methodology
A definition of the risk premium requires the specification of a temporal perspective.

This leads to two different, necessarily to distinguish, definitions. The first is known
as the ex ante or expected risk premium, and the second as the ex post or realized
risk premium.
To define the risk premium I will use the following notation in this dissertation: π
stands for the risk premium, S(t) for the spot price at time t, and F(t,T) for the
futures price at time t for a future with delivery in T.13 Et equals the expectation
13
Please note that T is in the case of electricity futures always a time period rather than a
point-in-time.
128
operator at time t. Only information that is available up to this time is included in

the expectations.
The ex ante or expected risk premium π(t, T ) at time t in a future F(t, T) with
delivery in T is defined as
π(t, T ) = F (t, T ) − Et [S(T )]. (4.1)
The second part of the right side of equation (4.1), the unobservable expected future
spot price, Et [S(T )], is of critical importance for the estimation of the ex ante risk
premium. Empirical research on the ex ante risk premium always requires a specifi-
cation of a spot price forecast model to estimate the expected spot price. The choice
of an appropriate spot price model is essential for the estimation of the ex ante risk
premium. However, spot price models are very sensitive to the specific underlying
assumptions.14 Thus, consistent and robust results for the ex ante risk premium are
difficult to obtain. Consequently, the focus of the empirical literature is on the ex
post risk premium.
The ex post or realized risk premium π(T ) is defined as
π(T ) = F (t, T ) − S(T ). (4.2)
The notation of the risk premium in equation (4.2) signals that the observation takes
place at maturity of the future in T. The main advantage of this definition is the
availability of all relevant data in the estimation procedure.
Definition (4.1) and (4.2) can be linked through equalizing. This results in
π(t, T ) − S(T ) = F (t, T ) − Et [S(T )] − S(T ). (4.3)
Under the assumption that market participants form their forecasts based on rational
expectations15 , equation (4.3) can be written as
F (t, T ) − S(T ) = π(t, T ) + t . (4.4)
According to equation (4.4), the ex post risk premium equals the ex ante risk pre-
mium plus a noise term. As the market participants form their expectations ratio-
14
Cf. Karakatsani & Bunn (2005) for a discussion of the difficulties regarding the application of
spot price models.
15
The assumption of rational expectations implies that (i) expectations are unbiased, i.e. the
forecast error equals zero: E(t ) = 0, (ii) forecast errors are uncorrelated, i.e. forecast errors in the
past contain no information regarding an improvement of the forecast, and (iii) expectations are
complete, i.e. the forecast cannot be improved based on the present information; cf. Schnorrenberg
(2006), pp. 157-158.
129
nally, it is assumed that the resulting average forecast error is zero.16

The assumption of an average forecast error of zero is strong. In particular for
a young market with a low number of market participants trading a commodity
such as electricity with all its special characteristics. Thus, it has to be noted that
an interpretation of the ex post risk premia is always problematic because of this
assumption.
A direct implication of this assumption is the question whether a certain group
of market participants possesses superior forecasting abilities. Within the applied
methodology the estimated risk premia equal the profit of either the long or short side
in the market. Superior forecasting abilities would result in a mix of the profitability;
the risk premia would not be directly distinguishable from forecast profits and the
results from an ex post risk premia analysis would be biased. It is even possible that
the estimated profits are completely the result of superior forecasting ability and
not of the earning of risk premia. Parts of the academic literature find support for
superior forecasting abilities, at least in certain markets.17 However, in accordance
with the broad literature I presume that the assumption of rational expectations
holds and hence no superior forecast profits occur.
Bryant et al. (2006) conclude the assumptions and problems underlying an empirical
analysis of risk premia in futures markets concisely by stating the following:
In summary, risk premiums that may exist in futures market cannot

be observed, because the expected future cash price cannot be observed.
The standard empirical practice then is to check for speculative profits,
which would be consistent with the existence of risk premiums. If specula-
tive profit exist (the evidence on this is mixed), they must be decomposed
into profits due to forecasting ability, which is also unobserved, and any
residual profits. The existence of residual profits is interpreted as evi-
dence that risk premiums are present. These premiums may be due to
systematic risk if futures price changes are correlated with returns to to-
tal wealth. After adjusting “observed” risk premiums for systematic risk,
it is then inferred that any residual risk premium that is not due to sys-
tematic risk may be due to hedging pressure, if measures of these two
phenomena are correlated. The path by which a researcher might find
16
Cf. section 2.3.2.2 for a discussion of potential interpretations of the ex post risk premium.
17
Chang (1985) examines the profits to speculators in wheat, corn, and soybeans futures markets.
The author finds evidence that large wheat speculators seem to possess some superior forecasting
ability over the period 1951 to 1980. Their profits are hence the sum of a risk premium and a reward
for their forecasting abilities. However, Chang (1985) makes no attempt to quantify the relative size
of the two components. Leuthold et al. (1994) analyze the frozen pork bellies market on the Chicago
Mercantile Exchange. When analyzing a dataset from 1982 to 1990 with daily trading activities the
authors find evidence that a group of traders seems to be able to generate superior price forecasts.
130
evidence consistent with the generalized theory of normal backwardation

is so convoluted it is littler wonder that no consensus may been reached.18
For empirical purposes I will calculate the ex post risk premium as
T
1X
π(T ) = (F (t, T ) − S(T )). (4.5)
T t=1
The spot price S(T) in equation (4.5) is calculated as the average of the hourly prices
during the delivery period
n
1X
S(T ) = Si (t) (4.6)
n i=1
with n being the number of hours during the delivery period.

In addition I calculate a relative risk premium, πrel , defined as
T
1X F (t, T ) − S(T )
πrel (T ) = ( ). (4.7)
T t=1 F (t, T )
The relative risk premium can be interpreted as the percentage of the futures price
which is paid due to hedging purposes.
4.4 Empirical Results
4.4.1 Market Liquidity
Considering exchange-trading, market liquidity is always a critical issue, in particular

in newly deregulated markets as the electricity market. Together with transaction
costs and transparency standards liquidity is usually used as a measure for the func-
tioning of an exchange.19 Furthermore, a public discussion on the efficiency of the
trading process has accompanied electricity exchanges since the beginning. A high
level of liquidity (together with a high number of market participants) is of course
the best argument to face this discussion.
4.4.1.1 Liquidity Spot Market
In 2009 the traded volume in the spot market amounted to 203 TWh. Similar to other
electricity exchanges most of the traded volume in the spot market is observed in
18
Bryant et al. (2006), p. 1043.
19
Cf. Wawer (2007), p. 25.
131
Figure 4.1
Traded Volume Spot Market 2003-2009
Traded volume in the spot market over the period 2003 to 2009. All market segments of the intraday market are included.
In 2009 for the first time electricity with delivery in the French market area is included in the statistic.
200
160
Traded Volume [TWh]
120
80
40
0
2003 2004 2005 2006 2007 2008 2009
Source: Own work.
the day-ahead market. Regarding the electricity traded for delivery in Germany, the
traded volume in the day-ahead market was around 136 TWh. In the intraday market
approximately 6 TWh were traded. This relatively low liquidity in the intraday
market could be due to the fact that this market segment is mainly used as a balance
market for short-term adjustments.20 The evolution of the traded volume in the spot
market over the period 2003 to 2009 is depicted in figure 4.1.
A steady increase in the traded volume over the last years is observed. However,
the increase in traded volume in 2009 appears surprising against the background
of the economic crisis and the development in the futures market discussed below.
However, it can easily be explained by the fact that traded electricity for delivery
in France was counted for the first time in this year. This volume amounted to 53.6
TWh in 2009.21 Without this effect a slight decline in the traded volume would have
been observed.
From the beginning of the sample period the liquidity in the day-ahead market was
relatively high and has steadily developed over the last years. Regarding the traded
volume in the individual contracts a significantly higher volume is observed in the
peak hours on working days. On non-working days no clear daily structure – except
20
Cf. Weber (2010), p. 3.
21
Cf. EEX (2010b), p. 5.
132
Figure 4.2
Daily Number Contracts without Trading in the Intraday Market
Daily number of hour contracts without trade in the intraday market. Sample period: September 25, 2006 to June 30, 2010.
As the intraday market is a continuous market no trading implies no match between sell and buy orders.
24
18
# Contracts Without Trade
12
0
Sep-06 Mar-07 Sep-07 Mar-08 Sep-08 Mar-09 Sep-09 Mar-10
Source: Own work.
a light peak in the morning hours – is observed.

When examining the period in which the block contract market was active the vol-
ume in this submarket was relatively low. In 2006 and 2007, for example, the traded
volume was 0.86 TWh and 0.61 TWh, respectively. It stands out that on approx-
imately 30% of all days no trading took place in the block contract market. On
non-working days this number was even higher. In particular the peak contract with
delivery on non-working days has almost never been traded.
Trading in the intraday market started in September 2006. However, it took almost
one year until the liquidity reached a sufficient level. Not only a low total volume
but also a high number of hour contracts without trading, as displayed in figure 4.2,
is observed in particular in the first year of operation.
For the empirical analysis of the spot market in this chapter, I restrict the available
data for the spot market as follows: The day-ahead market dataset which is used
for the spot market analysis starts August 1, 2002 as the EEX changed the trading
system that day. The same applies to the block contract market data. However,
regarding the analysis of the futures market in section 4.4.2.2 I also use the July
2002 data from the day-ahead market to obtain one additional complete future price
time series, i.e. risk premia time series. I exclude the peak contract from the block
133
Figure 4.3
Traded Volume Futures and Forward Market 2003-2009
Traded volume in the futures (darker part of the bar) and forward market (lighter part of the bar) over the period 2003
to 2009. The forward market data only include the volume that was cleared over the EEX.
1400
1200
1000
Traded Volume [TWh]
800
600
400
200
0
2003 2004 2005 2006 2007 2008 2009
Source: Own work.
contract market due to its low liquidity. For the intraday market I decide to skip the
first 15 months of trading due to the low liquidity. This decision is in particular based
on the high number of hour contracts without trade over this period. Furthermore, I
assume that 15 months are a necessary time period to gain market knowledge by the
market participants. Thus, the first day of interest for the analysis of the intraday
market is January 1, 2008.
4.4.1.2 Liquidity Futures Market
The total traded volume in the futures market and forward market, as far as the OTC
trades were cleared by the EEX, was 1025 TWh in 2009. 740 TWh or approximately
72% of this volume is observed in the forward market and 285 TWh in the futures
market.22 Figure 4.3 contains the evolution of the traded volume in the futures and
in the forward market over the last seven years.
The bar‘s dark part in figure 4.3 reflects the exchange-traded volume, the light part
the volume observed in OTC clearing. As can be seen a volume decrease of around
10% is observed in 2009 due to the financial and economic crisis that erupted in
2008.
22
Cf. EEX (2010b), p. 13.
134
Figure 4.4
Number Traded Month Futures with Respect to Time-to-Delivery
Average number of daily traded contracts with respect to time-to-delivery. The month futures are synchronized according
to the delivery month. The straight line represents the month base futures, the dashed line the month peak futures.
240
160
# Contracts
80
0
0 30 60 90 120 150 180
Days-to-Delivery
Source: Own work, based on Pietz (2009a).
Taking a closer look at the open interest, in the fourth quarter of 2002 the daily
open interest in all futures contracts averaged to approximately 29 TWh. At the
end of the sample period, the second quarter of 2010, an average open interest of 530
TWh is observed. This represents an astonishing increase in the open interest in the
magnitude of a factor 18 in eight years and speaks for a liquid and a well developing
market. The increase in the open interest was smooth and took place along with an
increasing number of market participants, and of traded contracts.
Futures tradable at the beginning of the sample period were: a month future with
delivery during the trading month, month futures with a time-to-delivery of up to
six months, quarter futures for the next seven quarters, and year futures for the
next three years. To extend the term structure over the past years new futures were
introduced by the EEX .
When analyzing the number of traded contracts a typical pattern for futures markets
is observed. Trading mainly takes place in futures with a short time-to-delivery. The
average daily number of traded contracts in the month futures, both base and peak,
over the whole sample period is depicted in figure 4.4.
The highest liquidity is observed in the futures with a short time-to-delivery. This
135
is a feature which is shared with other commodity futures markets.23 Furthermore,

figure 4.4 reveals the low liquidity of month futures with a time-to-delivery longer
than three months. The same is observed for futures with a longer time-to-delivery,
i.e. quarter and year futures. Consequently, the question on the reliability of these
futures prices arises, as the regular trading in a futures contract appears to be a
necessary condition to ensure meaningful price information. However, I recall that
arbitrage opportunities align some of the futures prices and that prices for non traded
futures are determined by the EEX.24
Table 4.2 illustrates the importance of the setting of prices by the EEX for month
futures. As already indicated by figure 4.4, the percentage of days without trading
is significant for the month futures with a longer time-to-delivery. Moreover, the
liquidity of the peak futures seems to be significantly lower than the liquidity of the
month futures. However, the high percentage of days without trading in the month
futures with a longer time-to-delivery, in particular the five and six month future25 ,
not necessarily implies factious prices as the arbitrage relation to the quarter futures
secures a pricing around the ‘fair‘ market value.
For the empirical analysis of the futures market in this chapter, I restrict the avail-
able data for the futures market as follows: I decide to only analyze the month
futures as the liquidity of futures with a longer time-to-delivery appears as too low.
Furthermore, I exclude the data of the month futures during the delivery month; the
reasoning for this is explained in section 4.4.2.2.
4.4.2 Descriptive Results
4.4.2.1 Spot Market
In the following I report the descriptive statistics for the three market segments of
the spot market – the day-ahead market, the intraday market, and the former block
contract market – separately. I focus on the day-ahead market as it is the market
segment with the highest liquidity. In addition, it is the market segment which serves
as underlying for the futures market and as a reference market for the whole German
electricity market.
Day-Ahead Market
To obtain a first impression of the day-ahead market data, I display the daily price
23
Cf. Geman (2005), p. 20.
24
Cf. section 2.2.4.3 for a discussion of the price determination process for futures without trading
on a certain day.
25
Cf. section 4.4.2.2 for a discussion of the terminology.
136
Table 4.2
Percentage Days without Trading in Month Futures by Year
Percentage of days without trading in the month futures. All days between August 1, 2002 and June 30, 2010 are included. Both the
month base (first table) and peak (second table) future are analyzed. Values in %.
Month Base
Future 2002 2003 2004 2005 2006 2007 2008 2009 2010
One 19.05 19.84 14.12 5.88 3.97 2.00 1.98 2.77 2.38
Two 40.95 32.94 45.88 26.27 17.46 8.80 18.25 10.67 8.73
Three 48.57 49.60 65.88 51.76 36.11 22.00 36.11 37.15 18.25
Four 65.71 70.63 81.57 75.69 55.56 49.20 54.37 64.03 33.33
Five 90.48 82.14 86.27 78.82 71.83 69.60 64.29 77.87 51.59
Six 95.24 94.84 93.73 85.88 82.94 80.40 76.59 84.98 59.52
Month Peak
Future 2002 2003 2004 2005 2006 2007 2008 2009 2010
One 33.33 41.27 21.57 15.29 20.63 13.60 18.25 13.83 11.90
Two 56.19 57.54 40.39 38.82 42.06 31.60 42.06 39.92 34.13
Three 64.76 73.81 64.31 57.25 57.14 58.80 59.13 67.98 49.21
Four 92.38 88.49 80.39 78.04 75.00 77.20 81.75 81.42 61.11
Five 100.00 94.84 88.63 78.82 82.94 86.80 85.32 91.30 80.95
Six 100.00 98.02 92.94 85.88 85.32 91.20 92.06 95.65 88.10
Source: Own work.
during the sample period in figure 4.5. The daily price is calculated as the arithmetic
average of the 24 hourly prices; the EEX publishes this price as the daily Phelix Base.
In addition, to illustrate the stylized facts of electricity prices discussed in section
2.2.2.1, the absolute daily (logarithmic) returns of the time series shown above are
depicted in figure 4.6.
Figures 4.5 and 4.6 reveal the high volatility of the daily day-ahead market prices.
Moreover, the discussed frequent price spikes are clearly observed in the price time se-
ries. The highest daily price is observed on July 27, 2006 with 301.54 Euro. July 2006
was a month with persistent high temperatures which resulted in a high power de-
mand and a lower power plant output due to high river temperatures.26 Furthermore,
the illustration of the daily absolute returns in figure 4.6 allows the identification of
volatility clusters, a common observation in financial markets.
A near examination of the time series in figure 4.5 leads to the observation of two neg-
ative daily prices at the end of the sample period. This is surprising as the day-ahead
market is by far the market segment with the highest liquidity. In addition, the first
occurrence of daily negative prices takes place after eight years of trading. However,
this has to be seen against the background of at least three developments. First,
26
Cf. Viehmann (2009), p. 7.
137
Figure 4.5
Daily Prices Day-Ahead Market
Time series of daily prices on the day-ahead market. Sample period: August 1, 2002 to June 30, 2010. Both working and
non-working days are included. The daily price is calculated as an arithmetic average of the 24 hourly prices.
300
Price [Euro]
200
100
0
2002 2003 2004 2005 2006 2007 2008 2009 2010
Source: Own work, based on Pietz (2009b).
Figure 4.6
Daily Absolute Returns Day-Ahead Market
Time series of daily absolute returns on the day-ahead market. Sample period: August 2, 2002 to June 30, 2010.
Both working and non-working days are included.
250
200
Absolute Return [%]
150
100
50
0
2002 2003 2004 2005 2006 2007 2008 2009 2010
Source: Own work.

138
negative prices were introduced not until 2008 in the day-ahead market. Second,
the current economic crisis was so severe that a decrease in the electricity demand
of around 5% was observed in 2009 in the German market.27 Third, the regulatory
framework in Germany was changed. As the EEG28 aims to increase the ratio of
electricity generated based on renewable energies, the absorption of this electricity
is compulsory. Thus, the TSOs are forced to inject the electricity generated by these
technologies at the time of generation. Starting in 2010, the underlying mechanism
was changed and the whole electricity generated within the EEG is now sold in
electricity wholesale markets, i.e. the supply at the EEX is increased.29
Both observations of the negative daily prices are on non-working days. The first
on a Sunday, October 4, 2009, and the second on a Saturday, December 26, 2009.
The German Federal Ministry of Economics and Technology assigned an academic
study regarding the occurrence of daily negative prices. The results of the study,
conducted by Nicolosi et al. (2010), are that negative price spikes are caused by a low
load in combination with high supply in terms of electricity generated by renewable
energies. Thus, the daily negative price on the day in October 2009 which was a
non-working day is probably explained by a low demand in the morning hours30 and
a high electricity supply due to a high wind energy input.
Recalling that seasonality is one of the prevalent characteristics of electricity prices –
often observed on three time scales: on a daily, weekly, and monthly scale – I analyze
the day-ahead market data in terms of this characteristic. First, I am interested in
the weekly seasonality. Therefore, I calculate the average daily price on all weekdays
and find that the prices on the weekends are significantly lower. Then I go a step
further and use the official trading calendar of the EEX.31 I calculate the average
prices on working and non-working days. Non-working days are weekend days and
public holidays. The results show that prices on public holidays during the week
are also significantly lower than on normal working days. I hence exclude the public
holidays on weekdays from the analysis. As a final result, I get a maximum average
daily price – the daily Phelix Base – on Tuesdays (49.56 Euro), almost identical
prices on Wednesdays and Thursdays, and slightly lower prices on Mondays (47.05
Euro) and Fridays (45.13 Euro). In contrast, the average prices on Saturdays (35.42
Euro) and on Sundays (27.86 Euro) are significantly lower. The results of this weekly
27
Consumption data can be obtained from the European Network of Transmission System Op-
erators for Electricity (www.entsoe.eu) on various time scales.
28
Cf. section 2.2.3.3 for a discussion of the EEG.
29
Cf. Nicolosi et al. (2010) for a discussion of the mechanism and the recent change.
30
On October 4, 2009 only five hours with negative prices are observed. These are the hours 2
to 6. However, in particular the price in hour 3 which is around minus 500 Euro causes an average
daily negative price.
31
The official trading calendar of the EEX is available on the homepage (www.eex.com).
139
Figure 4.7
Daily Prices on a Weekly Scale Day-Ahead Market
The average daily price is calculated over the period August 1, 2002 to June 30, 2010. Non-working days on weekdays
are excluded. The daily price corresponds to the PHELIX Base.
60.00
40.00
Price [Euro/MWh]
20.00
0.00
Monday Tuesday Wednesday Thursday Friday Saturday Sunday
Source: Own work.
analysis are depicted in figure 4.7.

Based on the results above I decide to distinguish in the following between working
and non-working days for the day-ahead market and the two other market segments
of the spot market. Thus, all price observations from Saturdays, Sundays, and public
holidays on weekdays are clustered as Non-Working Days. All other price observa-
tions are clustered as Working Days. The descriptive statistics for the reordered
dataset with working and non-working days are reported in tables 4.3 and 4.4.
Analyzing the minimum prices, I count 28 hours with negative prices on working days
and 66 hours with negative prices on non-working days.32 As can be seen in table 4.3
the hours with negative price observations on working days are solely the morning
hours, namely in the time period between midnight and 6 am. On non-working days
the negative price hours are more mixed and even occur in the afternoon hours.
However, the majority of the observations is also observed in the morning hours.
The lowest price in the day-ahead market is observed on a non-working day in hour
3 with minus 500.02 Euro on the already above discussed October 4, 2009.
When examining the hourly prices in table 4.3 I detect the expected seasonality on
a daily basis. The daily seasonality on working days is depicted in figure 4.8 for a
32
Cf. Nicolosi (2010) for a discussion and analysis of these negative hourly prices.
140
Table 4.3
Descriptive Statistics for Hourly Day-Ahead Market Prices (Working Days)
Descriptive statistics for hourly day-ahead market prices on working days. The covered sample period is August 1, 2002 to June 30, 2010.
Hour 1 stands for the hour contract with delivery between 0 am and 1 am. The following hours are set accordingly.
Hour Mean Std.Dev. Minimum Median Maximum Skewness Kurtosis Hour Mean Std.Dev. Minimum Median Maximum Skewness Kurtosis
1 31.00 12.69 -16.67 28.79 75.01 0.79 0.70 13 60.84 32.56 14.28 53.04 699.81 6.34 94.60
2 27.10 12.89 -151.67 25.51 69.63 -0.83 18.60 14 58.79 31.23 14.63 50.93 699.88 6.44 103.36
3 24.39 12.20 -101.52 23.10 64.09 -0.27 6.72 15 55.71 31.06 13.04 47.96 800.09 8.34 170.28
4 22.17 12.34 -149.94 21.63 60.20 -1.61 24.38 16 52.12 27.62 11.76 45.01 693.23 7.65 150.23
5 23.22 11.69 -101.50 22.38 61.93 -0.70 12.34 17 50.60 24.60 15.23 43.53 300.01 2.82 15.99
6 29.52 11.60 -9.98 27.58 70.51 0.79 0.77 18 55.78 40.94 15.17 45.79 821.90 7.81 104.84
7 38.31 14.79 1.09 35.08 104.93 0.96 1.06 19 60.07 76.87 13.92 47.60 2437 20.88 572.75
8 53.71 24.34 3.03 46.53 301.01 2.35 13.76 20 53.20 27.07 14.54 44.96 300.01 2.55 13.09
9 57.66 28.20 8.02 50.11 437.26 3.33 26.09 21 49.49 20.50 15.81 43.64 194.62 1.30 2.34
10 60.04 30.78 13.39 51.91 499.68 3.67 30.35 22 43.37 16.14 12.48 39.57 118.93 1.08 1.09
11 62.65 37.97 13.61 53.92 998.24 9.41 195.69 23 40.94 14.83 13.93 37.71 94.82 0.95 0.49
12 70.71 65.39 15.54 59.06 2000 17.76 466.73 24 34.21 12.43 9.15 32.13 80.98 0.90 0.64
ALL 46.48 33.98 -151.67 40.09 2437 18.51 965.20
Table 4.4
Descriptive Statistics for Hourly Day-Ahead Market Prices (Non-Working Days)
Descriptive statistics for hourly day-ahead market prices on non-working days. The covered sample period is August 1, 2002 to June 30, 2010.
Hour 1 stands for the hour contract with delivery between 0 am and 1 am. The following hours are set accordingly.
1 30.30 14.28 -119.90 28.30 76.02 -0.60 13.53 13 38.32 15.77 1.88 36.09 99.82 0.97 1.11
2 25.18 14.63 -119.96 23.99 71.07 -1.33 17.42 14 33.65 14.27 -0.04 32.00 91.44 0.86 0.95
3 21.99 22.31 -500.02 21.57 67.93 -14.58 338.59 15 30.14 13.38 -0.74 28.88 87.16 0.78 0.89
4 20.16 13.67 -119.98 19.09 69.52 -1.41 18.33 16 28.16 12.94 -1.93 26.86 82.93 0.75 0.84
5 19.11 13.03 -119.97 17.92 69.92 -0.95 15.67 17 28.76 13.28 0.00 26.90 80.08 0.82 0.89
6 19.15 13.08 -119.98 18.12 69.90 -0.72 14.28 18 34.37 16.07 0.00 31.46 109.95 1.00 1.12
7 15.69 15.34 -199.99 14.52 75.25 -3.31 48.58 19 39.27 17.78 0.26 35.39 119.98 0.93 0.68
8 20.78 15.75 -199.94 19.85 80.68 -2.55 43.12 20 39.86 17.51 1.01 36.05 111.02 0.91 0.77
9 27.08 15.37 -119.96 25.56 91.05 -0.21 9.90 21 37.24 14.88 2.27 34.95 105.31 0.96 1.39
10 33.33 15.89 -36.10 30.91 96.04 0.87 1.62 22 34.82 13.34 2.23 33.40 89.96 0.84 0.99
11 36.91 16.25 0.00 34.38 99.75 1.04 1.43 23 36.56 13.98 4.53 34.81 87.26 0.84 0.64
12 40.36 16.28 1.92 37.66 105.02 1.04 1.27 24 29.94 12.40 -32.80 28.30 75.09 0.57 1.54
ALL 30.05 16.87 -500.02 28.09 119.98 -1.28 51.51

141
Figure 4.8
Hourly Prices on a Daily Scale Day-Ahead Market
The average hourly price is calculated over the period August 1, 2002 to June 30, 2010. Non-working days (also on
weekdays) are excluded.
80
60
Price [Euro/MWh]
40
20
0
1 3 5 7 9 11 13 15 17 19 21 23
Source: Own work.
better illustration.
On average, the hourly prices on working days are three times higher around midday
compared to the early morning hours. A similar pattern is observed for the volatil-
ity: When surveying the maximum prices and the third and fourth moment of the
distributions I detect hours with frequent price observations in the magnitude of ten
times higher than the average price. The dataset even includes four observations
of prices above 1,000 Euro: Two in hour 12 (11 am – 12am) on July 25, 2006 and
July 27, 2006; two in hour 19 (18 pm – 19 pm) on January 07, 2003 and November
11, 2006. The last observation will be of importance at a later point in this chapter
when I discuss the volatility of the estimated risk premia. This one outlier causes a
significant bias in the corresponding descriptive statistics for this hour.33
The observed maximum prices are the well-known and dreaded price spikes or jumps34 ,
one of the unique characteristics of electricity prices. The price jumps combined with
the high skewness of the price distributions during high demand hours underline the
importance of hedging in electricity markets.35 The two peaks in the average prices,
33
Cf. Viehmann (2009) for a discussion of the factors causing these maximum prices in the
German market.
34
Cf. for example Seifert & Uhrig-Homburg (2007) for a discussion of price spikes in electricity
markets.
35
Cf. for example Deng & Oren (2006) for an overview on hedging in electricity markets.
142
Figure 4.9
Hourly Demand in the German Electricity Market
Hourly load on June 18, 2008 (light bar) and on December 17, 2008 (dark bar) in the German market. The data are
obtained from the European Network of Transmission System Operators for Electricity (ENTSOE).
100
80
Hourly Load [GW]
60
40
20
0
1 3 5 7 9 11 13 15 17 19 21 23
Source: Own work.
the one in hour 12, the other in hour 19 deserve a near discussion. According to
Viehmann (2009) I find that peak prices around the hour 12 contract occur in sum-
mer months and around the hour 19 contract mainly in winter months.
The price spikes are explained by different demand patterns in summer and winter
months. One summer and one winter day are shown, as an example, in figure
4.9. The light bar in the figure depicts the hourly load on June 18, 2008, the dark
bar on December 17, 2008. The data are obtained from the European Network of
Transmission System Operators for Electricity (ENTSOE). As can be seen the price
spikes observed above overlap with peaks in the hourly load.
Coming back to the descriptive statistics in table 4.3, it also stands out that, when
comparing working to non-working days, not one individual price spike is observed
on a non-working day. The volatility and the skewness of the distributions on non-
working days are also significantly lower than on working days. To further analyze
this, I order the price data by years, both for the base and peak hours. The descrip-
tive statistics for these data can be found in table 4.5.
The average prices in table 4.5 seem to incorporate a positive drift over the period
2002 to 2008. Except for the year 2007 I observe a steady increase in the average
price. A particularly strong increase occurred in 2005 when the EU ETS was in-
143
Table 4.5
Descriptive Statistics for Daily Day-Ahead Market Prices by Year
Descriptive statistics for daily day-ahead market prices by year, both for the base and peak hours. Sample period: August 1, 2002 to
June 30, 2010. The daily price is calculated as an arithmetic average of the 24 hourly prices.
Base Peak
Year Mean Std.Dev. Minimum Median Maximum Skewness Kurtosis Mean Std.Dev. Minimum Median Maximum Skewness Kurtosis
2002 22.98 8.22 3.47 22.89 49.77 0.19 0.09 29.36 11.33 4.14 28.88 75.40 0.47 1.05
2003 29.49 13.07 3.12 29.05 163.46 3.78 32.58 37.00 21.44 0.80 35.24 277.64 5.21 48.71
2004 28.52 6.53 12.06 29.38 46.61 -0.37 0.03 33.99 8.56 11.79 35.12 60.17 -0.13 0.00
2005 45.98 18.42 13.56 42.37 145.97 2.47 9.15 56.00 28.55 16.03 49.62 226.33 2.97 12.12
2006 50.79 24.50 13.98 46.86 301.54 4.42 36.69 63.81 40.98 17.42 57.74 543.72 6.45 63.49
2007 37.99 19.90 5.80 32.71 158.97 2.38 7.98 48.75 30.76 6.76 41.00 248.38 2.77 11.11
2008 65.76 18.12 21.03 65.71 131.40 0.26 0.21 79.43 24.24 21.54 76.86 177.49 0.56 0.47
2009 38.85 11.86 -35.57 37.92 86.36 -0.15 5.84 46.83 14.04 9.47 44.24 114.63 1.10 2.38
2010 41.12 7.13 16.69 41.58 54.07 -0.73 0.86 46.50 9.17 18.19 46.60 65.81 -0.52 0.70
troduced and emission rights were incorporated as a direct production factor in the
electricity generation process. The adoption of the emission’s right price as a cost
factor in electricity generation also represents a potential source for the increasing
volatility.36
In 2009, a drop of the price level in the magnitude of 40% is to observe. This
is probably due to the beginning of the current economic crisis, starting with the
collapse of the investment bank Lehmann Brothers in September 2008. Similar to
other markets, this was the point-in-time when prices and volume started to decrease
in the German electricity wholesale market. However, the significant price decrease
is observed in the first quarter 2009. The prices remain at the then reached price
level until the end of the sample period.
On average, prices in peak hours are around 20% higher than in base hours. This
ratio is quite stable over the whole sample period.37 The volatility in the peak hours
is also significantly higher compared with the base hours.
The last question of interest for the day-ahead market is a potential yearly season-
ality. The average month prices are shown in figure 4.10. A maximum price in July,
October, and November is observed. May and August seem to be the months with
the lowest price.
36
Cf. Zachmann & von Hirschhausen (2008) for a discussion of the mechanism behind this
observation.
37
The sample period August 1, 2002 to June 30, 2010 contains 2,891 daily Phelix Base and Phelix
Peak prices. Around 75% of the peak prices are between 10% and 30% above the base price; around
90% of the peak prices are between 5% and 35% above the base prices.
144
Figure 4.10
Monthly Prices on a Yearly Scale Day-Ahead Market
The average monthly price is calculated over the period August 1, 2002 to June 30, 2010. Non-working days on
weekdays are excluded. Shown are the montly base prices.
50.00
40.00
Price [Euro/MWh]
30.00
20.00
10.00
0.00
ril
ay
ly
y
ry
er
ch
ne
st
er
r
be
be
ar
Ju
Ap
gu
ua
ob
eb
M
ar
Ju
nu
em
em
Au
br
ct
em
Ja
O
Fe
ov
pt
ec
Se
D
Source: Own work.
However, the results are mixed. When the monthly seasonality is analyzed year by
year, it seems that the price peak in summer months is not stable. In 2004, for
example, no peak prices in the summer months are observed. On the other side, a
clear price peak is observed in July 2007. In 2008, the prices are significantly higher
from October to December. A price peak occurs in September and October 2009.
Thus, a stable tendency for higher prices around October can be concluded. The
prices for summer months, i.e. around July, need to be observed in the coming years
to reach a conclusion on the yearly seasonality.

Table 4.6 reports the descriptive statistics for the block contract prices. In addition,
certain day-ahead market prices are included in the table as well.
The reported data for the day-ahead market in table 4.6 are the corresponding syn-
thetic block contracts.38 When comparing the block contracts with the synthetic
block contracts I observe that the descriptive statistics are very similar except for a
higher skewness and kurtosis of the day-ahead market data. This is probably due to
38
A synthetic block contract was constructed by bidding for the hour contracts in the day-ahead
market which spanned over the time period as covered by the corresponding block contract.
145
Table 4.6
Descriptive Statistics for Block Contract Market Prices
Descriptive statistics for block contract market prices (first table) and the corresponding (synthetic) block contract prices in the day-ahead market
(second table). Sample period: August 1, 2002 to August 31, 2008. The weekend base contract series starts in November 2002.
Block Contracts
Working Days Non-Working Days
Mean Std.Dev. Minimum Median Maximum Skewness Kurtosis Mean Std.Dev. Minimum Median Maximum Skewness Kurtosis
Base 47.70 22.56 12.00 40.31 179.33 1.61 4.25 28.26 14.32 5.25 24.23 66.50 0.89 -0.05
Peak 61.04 31.75 16.67 51.07 288.42 2.33 9.11 23.43 11.23 8.00 25.27 42.21 0.01 -0.93
Weekend
- - - - - - - 30.56 13.23 10.68 24.75 66.88 0.69 -0.76
Base
Day-Ahead Market Contracts
Working Days Non-Working Days
Mean Std.Dev. Minimum Median Maximum Skewness Kurtosis Mean Std.Dev. Minimum Median Maximum Skewness Kurtosis
Base 45.29 21.22 12.40 38.98 301.54 2.83 19.63 28.43 12.56 3.12 25.54 80.42 0.94 0.71
Peak 57.28 32.00 15.86 48.41 543.72 4.71 47.92 32.29 14.25 0.80 28.97 86.69 0.96 0.72
Weekend
- - - - - - - 29.56 11.63 9.24 26.04 70.95 0.87 0.14
Base
the frequent price spikes in this market.

I note that the high number and the uneven distribution of the non-trading days in
the block contract market have to be taken into account when the results obtained in
the following are interpreted. Another systematic pattern in the data, also already
observed by Ronn & Wimschulte (2009), is a higher number of days without trading
in contracts with delivery on Mondays.
Intraday Market
Before providing descriptive statistics for the intraday market, I have to decide which
price time series is appropriate for the estimation of the risk premia – either the last
price or the average price. The last price is the price of the last trade in an hour
contract; the average price the average price of all trades in an hour contract. There
is no information on the number and timing of individual trades in a certain hour
contract in the intraday market available. Thus, the only secure information on these
two prices is that the last price is, in the case that more than one trade in an hour
contract took place, chronological later than the average price. Comparing the two
time series with regard to systematic differences no obvious results are found. The
mean of the average price time series is slightly higher than the one of the last price
146
Table 4.7
Descriptive Statistics for Hourly Intraday Market Prices (Working Days)
Descriptive statistics for hourly intraday market prices on working days. The price of the last trade in the intraday market is used. Sample
period: September 25, 2006 to June 30, 2010. Overall 20,837 price observations on working days are included.
1 35.17 18.87 -200.00 34.00 84.00 -4.61 57.19 13 65.62 27.44 16.00 60.00 220.00 1.33 2.97
2 31.01 17.28 -200.00 30.00 75.00 -3.09 41.75 14 62.85 26.60 15.00 58.00 210.00 1.35 3.01
3 27.89 17.83 -200.00 28.00 79.00 -3.18 37.95 15 59.78 26.41 5.00 55.00 250.00 1.41 4.16
4 24.87 19.12 -200.00 25.00 70.00 -4.49 49.93 16 57.15 24.63 10.00 52.00 210.00 1.32 3.15
5 26.82 16.86 -200.00 27.00 76.00 -3.06 41.46 17 57.16 26.82 7.00 51.00 300.00 1.90 9.16
6 32.73 18.06 -200.00 32.00 84.00 -2.93 36.56 18 64.31 42.03 15.00 58.00 500.00 4.84 37.69
7 47.73 18.98 -10.00 45.50 111.00 0.41 -0.12 19 67.57 42.07 5.00 60.00 500.00 4.07 30.03
8 63.76 27.11 12.00 60.00 330.00 1.81 11.80 20 61.91 28.76 15.00 57.00 270.00 1.57 5.32
9 70.21 33.15 16.00 63.00 400.00 2.46 15.90 21 54.78 22.59 1.00 51.00 190.00 1.08 2.26
10 68.96 30.48 21.00 60.50 250.00 1.56 4.09 22 49.06 18.12 15.00 46.00 129.00 0.77 0.56
11 69.20 31.41 21.00 62.38 300.00 1.79 6.09 23 47.76 17.86 1.00 45.00 199.00 1.12 5.03
12 72.77 35.18 22.50 65.00 300.00 1.97 6.08 24 39.91 15.38 5.00 38.00 149.00 0.80 2.40
ALL 53.42 30.50 -200.00 48.00 500.00 2.10 18.95
time series; the volatility of the average price is slightly lower than the one of the
last price.
I believe that the last price is the one to be used for the following analysis. This is due
to the theoretical framework behind the use of the risk premia approach. I interpret
market segments of the spot market with earlier trading as futures markets. A day-
ahead market contract is, in general, interpreted as a future with a time-to-delivery of
one day. The time difference between trading in these two market segments is hence
the main characteristic of interest. The last price is the price observation which
maximizes the temporal difference between trading in the day-ahead and in the
intraday market. Thus, it is the price which should be used. However, considering
the thin trading in the intraday market and the presumably uneven distribution
of trades during the permitted trading phase for a specific contract, I conduct the
analysis also based on the average prices. I report potential differences in the results
in the following.
Descriptive statistics for hourly last prices on working and non-working days on the
intraday market are reported in table 4.7 and table 4.8, respectively. The intraday
market price data are similar to the day-ahead market price data.39 Perhaps the
39
The descriptive statistics for the intraday market cannot (directly) be compared with the de-
scriptive statistics for the day-ahead market as the sample period in the later case is much longer
and a strong increase in the prices was observed over the years.
147
Table 4.8
Descriptive Statistics for Hourly Intraday Market Prices (Non-Working Days)
Descriptive statistics for hourly intraday market prices on non-working days. The price of the last trade in the intraday market is used.
Sample period: September 25, 2006 to June 30, 2010. Overall 9,041 price observations on working days are included.
1 28.47 86.04 -1499 33.00 79.50 -17.00 301.72 13 42.89 16.85 1.00 39.35 101.00 0.76 0.41
2 25.45 56.67 -950.00 29.75 93.00 -15.49 264.51 14 37.81 16.61 -75.00 36.00 85.50 -0.37 5.04
3 23.40 49.86 -800.00 26.00 69.00 -14.12 228.56 15 32.60 22.26 -250.00 32.00 87.50 -5.73 67.71
4 22.76 25.70 -295.00 25.00 68.00 -7.24 81.88 16 30.25 42.45 -750.00 31.00 95.00 -15.23 277.70
5 22.39 19.51 -190.00 23.50 69.00 -3.92 41.07 17 32.69 20.57 -200.00 31.00 90.00 -3.54 40.49
6 22.44 18.59 -170.00 21.50 66.00 -3.10 32.07 18 41.46 19.11 -2.00 39.00 105.00 0.49 -0.08
7 20.91 23.67 -198.00 21.00 80.00 -3.83 30.99 19 47.64 21.78 6.00 45.00 142.00 0.83 1.20
8 26.22 22.94 -199.00 26.00 77.00 -3.92 35.63 20 47.64 21.27 -2.00 45.00 125.00 0.79 0.70
9 31.93 20.96 -190.00 32.00 95.00 -3.18 34.53 21 43.55 17.82 -1.00 41.00 113.00 0.64 0.39
10 40.05 17.97 1.00 38.00 105.00 0.62 0.54 22 40.30 17.46 -90.00 39.00 89.00 -0.88 7.55
11 42.74 17.02 6.00 40.00 102.00 0.76 0.68 23 42.33 16.96 1.00 40.00 96.00 0.44 -0.24
12 44.87 17.03 1.00 41.00 103.00 0.75 0.69 24 31.23 48.29 -800.00 33.00 89.00 -13.87 231.66
ALL 34.77 32.42 -1499 34.50 142.00 -20.83 787.73
most striking differences are the significantly lower maximum prices in the intraday
market and the consequential smaller skewness and kurtosis of the price distributions
for the intraday market.
Moreover, similar to the day-ahead market, negative prices are observed in the in-
traday market, both on working and non-working days. On working days I count 71
hours with a negative price over the sample period, on non-working days 113 hours.
The lowest price is observed on a non-working day with minus 1,499 Euro in hour
1. Similar to the day-ahead market, the majority of the negative price observations
occurs in the morning hours.
Figure 4.11 depicts the price time series of four selected hour contracts over the whole
sample period on working days. Appendix A contains the time series of all 24 hour
contracts on working days to illustrate the different characteristics in dependence of
the delivery hour. In addition, Appendix B contains the time series of all 24 hour
contracts of the day-ahead market on working days in the period January 2, 2008 to
June 30, 2010 to amplify the differences between these two market segments.
The time series of the hour 1, 7, 13, and 19 contracts on working days are shown in
figure 4.11. It is to consider that the y-axis is solely identical for the hour 13 and
hour 19 contract. The time series of the hourly contracts show both different price
levels and volatilities. The time series of the hour 1 contract is, for example, quite
148
Figure 4.11
Selected Hourly Price Time Series Day-Ahead Market
Time series of hourly price over the period January 2, 2008 to June 30, 2010. The price time series contain price observations only for working days.
Missing values are due to non trading in the paricular contract.
Hour 1 Hour 7
50.00
150.00
Price [Euro/MWH]
Price [Euro/MWH]
-50.00
50.00
-150.00
-250.00 -50.00
Hour 19
Hour 13
400.00
400.00
300.00
300.00 Price [Euro/MWH]
Price [Euro/MWH]
200.00
200.00
100.00 100.00
0.00 0.00
Source: Own work.
smooth at the beginning of trading in the intraday market and then exhibits two
negative prices in 2009. The hour 19 contract on the other side is characterized by
a high volatility and the regular occurrence of price spikes. Finally, the hour 7 and
hour 13 contracts reveal a relatively smooth timely evolution.
The EEX provides no daily prices from the intraday market; this is in opposite to the
day-ahead market for which the daily Phelix Base and Peak is published. To allow a
comparison of these two market segments, I calculate a daily price for the intraday
market similar to the daily Phelix Base. I calculate the daily price as an average
of all hourly prices within a day. However, due to the low liquidity in the intraday
market in particular at the beginning of trading, I have to deal with days with less
than 24 hourly prices. To overcome this data problem I decide to fill gaps in the
dataset, i.e. hours without an observed price, with the next observed price in the
same hour contract. Thereby, I distinguish between working and non-working days.
Thus, for example, a gap in the dataset for the hour 7 contract with delivery on a
Saturday is filled with the next observed price in this hour contract on a non-working
day. The resulting daily price is shown in figure 4.12.
Compared with the daily prices in the day-ahead market – as shown in figure 4.5 –
it first stands out that no price spikes occur in the intraday market, at least not on a
149
Figure 4.12
Daily Prices Intraday Market
Time series of daily price on the intraday market. Sample period: July 1, 2007 to June 30, 2010. Both working and
non-working days are included. In the case of a missing price the price of the next trade in the same contract is used.
200
Price [Euro/MWh]
100
0
2008 2009 2010
Source: Own work.
daily scale. Second, similar to the two negative price observations in the day-ahead
market, negative daily prices are observed in the intraday market. In the second half
of 2009 five daily negative prices are observed. Third, beginning with the financial
and economic crisis in 2008, after a period of high volatility, a decrease in both price
level and volatility is observed.
4.4.2.2 Futures Market
The analysis of the futures market excludes the month future with the shortest time-
to-delivery due to its trading in the delivery month. The settlement of a cash-settled
future consists in the payment of the difference between the price at opening the
position and the realized average spot price during the delivery period.40 Thus,
trading in the delivery period effectively leads to a conversion to a future with a
shorter delivery period. This results in a lower volatility and a convergence of the
futures price to the average spot price. This effect is shown in figure 4.13.
Three time series are displayed in figure 4.13. The first, shown as a continuous
dashed line, is the daily day-ahead market price. The second, which is recognizable
through the weekend interruptions and a smooth evolution, is the daily price of the
40
The daily mark-to-market mechanism is ignored in this analysis.
150
Figure 4.13
Month Base Future and Underlying Spot Price
The continuous dashed line is the daily day-ahead market price. The time series with the weekend interruptions is the
daily price of the month base future. The solid line is the average day-ahead month price during the delivery month.
60
40
Price [Euro/MWh]
20
0
Jul-02 Aug-02 Sep-02 Oct-02 Nov-02 Dec-02 Jan-03
Source: Own work.
month base future with delivery in July 2003. Finally, the average monthly price
during the delivery month is displayed as a solid line, i.e. the number of data points
used for this calculation increases with the time. The y-axis is set between 0 and 60
Euro to illustrate the time series. Thereby, one data point, a daily price of 163.46
Euro on July 1, 2003, is not shown. It can be seen that the futures price converges
to the average monthly price at the end of the trading period of the future.
The terminology which is employed for the month futures can be clarified through
the example of the futures price series in figure 4.13. The shown future is the month
base future with delivery in January 2003. This future was traded between July
1, 2002 and January 31, 2003. I handle the price data of this future as follows:
During trading in July 2002 this future is termed as six month future, in August as
five month future, and so on. Finally, in December 2002, I term this future as one
month future. January 2003 data are excluded due to the problems discussed above
regarding futures being traded in their delivery period.
The final dataset comprises 90 month futures observed over their whole trading pe-
riod. They are characterized by their delivery month, e.g. January 2003. Considering
the definition of the ex post risk premium and the problem of separating forecast er-
rors and risk premia the low number of contracts is identified as a potential problem
151
Table 4.9
Descriptive Statistics Month Base Futures
The first table contains descriptive statistics on the price data, the second on the return data (computed as
log returns). The data are monthly (daily data in brackets). Price data in Euro, return data in %.
Future Mean Std.Dev. Minimum Median Maximum Skewness Kurtosis
One 43.6 15.4 21.31 38.93 89.46 0.87 0.19

(43.59) (15.57) (20.8) (39.2) (98.41) (0.9) (0.27)
Two 44.48 15.76 22.33 41.29 92.79 0.81 0.11
(44.46) (15.87) (21.48) (41.01) (96.76) (0.85) (0.17)
Three 44.95 16.15 21.68 41.09 91.75 0.93 0.51
(45.03) (16.26) (21.22) (41.21) (98.23) (0.95) (0.56)
Four 45.14 16.17 21.18 42.22 95.44 1 0.98
(45.24) (16.28) (20.8) (42.11) (101.94) (1.03) (1.01)
Five 45.45 16.21 21.3 44.02 96 0.95 1.07
(45.54) (16.31) (20.93) (44) (101) (0.97) (1.07)
Six 45.54 15.9 20.5 43.44 95.32 0.86 0.93
(45.61) (15.98) (20.35) (43.53) (102.75) (0.89) (0.96)
One 0.84 13.52 -33.7 2.77 43.26 -0.11 0.44

(0.04) (3.15) (-15.47) (-0.08) (27.58) (1.73) (15.21)
Two 0.71 13.13 -38.48 1.42 32.82 -0.14 0.16
(0.03) (2.66) (-18.05) (0) (24.83) (0.89) (17.22)
Three 0.79 13.17 -35.68 1.36 43.93 0 0.73
(0.04) (2.55) (-23.91) (0) (32.18) (1.86) (34.72)
Four 0.72 11.14 -34.73 1.75 31.46 -0.31 0.86
(0.03) (2.43) (-19.97) (0) (33.67) (1.29) (39.34)
Five 0.83 11.36 -28.22 1 27.86 -0.11 -0.06
(0.04) (2.39) (-22.25) (0) (30.75) (1.5) (36.66)
Six 0.71 10.86 -34.49 1.48 24.16 -0.48 0.48
(0.03) (2.38) (-25.7) (0) (27.84) (0.07) (36.51)
for the interpretation of the later results and their robustness.

Using the terminology introduced above, I report in table 4.9 and table 4.10 the
descriptive statistics on the month futures, both for the base and the peak future.
The upper part of the table contains the descriptive statistics on the price data with
monthly frequency. The lower part of the table contains the descriptive statistics
on the return data. The returns are calculated as log returns. The corresponding
values for data with daily frequency are reported in brackets. The monthly data are
calculated as the arithmetic average of all prices within one month.
Comparing the daily and monthly prices a smoothing is observed. This was ex-
pected due to the lower sensitivity of monthly prices to price spikes. The average
price increases between the first and the third month future. Thereafter, it remains
constant. The prices for peak futures are on average 18 to 19 Euro or around 41%
higher than the base future prices. I observe a decreasing volatility with increasing
time-to-delivery. Without further examination, I conclude that this may be inter-
preted as the Samuelson Effect.41
41
The Samuelson Effect (Samuelson (1965)) indicates that the volatility of futures prices de-
creases as time-to-delivery increases. This is explained by a lower sensitivity of long-term futures to
information inflow due to a longer remaining adjusting period.
152
Table 4.10
Descriptive Statistics Month Peak Futures
The first table contains descriptive statistics on the price data, the second on the return data (computed as
log returns). The data are monthly (daily data in brackets). Price data in Euro, return data in %.
Future Mean Std.Dev. Minimum Median Maximum Skewness Kurtosis
One 61.15 21.89 32.49 54.34 130.77 0.95 0.25

(61.17) (22.33) (31.55) (54.34) (141.56) (1.01) (0.39)
Two 62.89 22.49 33.34 56.89 131.4 0.87 0
(62.87) (22.73) (32.14) (56.63) (136.91) (0.91) (0.11)
Three 63.55 22.78 31.9 58.79 130.11 1 0.53
(63.67) (22.99) (31.5) (58.25) (139) (1.03) (0.57)
Four 63.86 22.51 31.24 60.7 131.08 1.09 1.07
(64.01) (22.68) (30.68) (60.29) (143) (1.11) (1.07)
Five 64.41 22.27 31.21 61.34 131.3 1 0.99
(64.54) (22.42) (30.88) (61.13) (143) (1.01) (0.96)
Six 64.61 21.91 30.61 61.68 134.19 0.95 0.96
(64.69) (22.05) (30.25) (60.32) (144.1) (0.96) (0.95)
One 0.64 17.15 -44.87 2.14 51.38 -0.06 0.61

(0.03) (3.99) (-22.63) (-0.12) (43.72) (1.86) (20.04)
Two 0.55 15.68 -47.23 0.97 49.45 0.08 0.95
(0.03) (3.11) (-21.31) (-0.02) (39.59) (1.94) (26.22)
Three 0.62 14.57 -39.01 1.03 35.53 -0.23 0.1
(0.03) (2.82) (-31.55) (0.01) (35.82) (0.94) (36.7)
Four 0.58 12.03 -38.11 1.41 27.22 -0.58 0.48
(0.03) (2.71) (-29.88) (0) (25.31) (0.46) (36.38)
Five 0.73 12.14 -31.37 0.9 25.6 -0.14 -0.18
(0.04) (2.55) (-28.95) (0) (26.29) (0.61) (32.4)
Six 0.61 11.46 -37.85 1.66 23.39 -0.56 0.71
(0.03) (2.32) (-25.28) (0) (19.45) (-0.83) (31.62)
The observed high maximum values in futures prices are unexpected, in particular
when compared to the realized monthly prices on the day-ahead market serving
as underlying of the futures market as shown in figure 4.14. It is apparent that
the maximum future prices are higher than the highest realized prices on the spot
market. In addition, the positive skewness suggests that several observations were
taken in this price region. There also seems to be a tendency for a co-movement of
spot and futures prices which results in a high correlation between the time series.
For a further analysis of this behavior I run a regression of the futures prices on
the spot prices. When doing so, I take into consideration that a regression of two
time series is only meaningful when both time series are stationary or cointegrated.
Otherwise misleading results could be obtained due to spurious regression. When
testing for unit roots in the time series using the Dickey-Fuller-Test, I find that the
null hypothesis (existence of a unit root) cannot be rejected. Tests for cointegration
deliver mixed results. I hence drive a regression with first differences and find a
relationship between the spot and the futures prices.42
The above results are also found in a recent work by Redl et al. (2009) who analyze
the price formation in the futures markets of the EEX and of the Nord Pool. The
42
However, the results are mixed and neither easy to interpret. Due to space considerations and
the work of Redl et al. (2009) discussed below I do not report the results here.
153
Figure 4.14
Monthly Price Day-Ahead Market
Monthly prices on the day-ahead market. The monthly prices are calculated as the arithmetic average of the hourly prices
within the month. The straight line represents the monthly Phelix Base, the dashed line the monthly Phelix Peak.
120
Price [Euro/MWh]
80
40
0
Jul-02 Oct-03 Jan-05 Apr-06 Jul-07 Oct-08 Jan-10
Source: Own work.
authors find that fundamental expectations or risk considerations cannot fully ex-
plain the difference between futures and spot prices. They conclude that an adaptive
price formation effect is apparently existent in both markets. This is interpreted as
evidence for the existence of systematic forecast errors. These findings question the
assumption of rational expectations underlying every analysis of risk premia from
an ex post perspective. However, the results have to be seen as primarily and it
is possible that, for example, market power is responsible for this effect. Further
research on this topic is definitely necessary.
4.4.3 Risk Premia in the German Electricity Market
4.4.3.1 Risk Premia in Spot Contracts
The results of the empirical analysis of the spot market are reported separately for
the block contract market and the day-ahead market. Furthermore, a potential term
structure of risk premia and potential drivers of the risk premia are discussed.
Risk Premia in the Block Contract Market

The risk premia in block contracts are estimated as the price difference between the
154
Table 4.11
Risk Premia in Block Contracts
Risk premia in block contracts on working days. The absolute and relative risk premia are reported. ***, ** and *
indicate significance at the 1%, 5% and 10% level; the Newey-West estimator is used to obtain robust standard errors.
Absolute Risk Premia Relative Risk Premia
Mean Median Std.Dev. Mean Median Std.Dev.
Base 0.79** 0.64 11.34 1.07** 1.35 17.12
Monday 2.11*** 1.41 6.19 4.25*** 3.29 10.97

Tuesday 0.03 0.81 16.69 -0.9 1.47 28.45
Wednesday 1.12* 0.69 9.57 1.34* 1.53 12.65
Thursday -0.17 0.27 11.52 -0.03 0.56 13.96
Friday 1.21* 0.40 8.96 1.47 1.04 11.84
Peak 1.55** 1.51 20.37 1.9*** 2.90 22.93
Monday 3.41*** 2.00 11.60 4.87*** 4.20 12.87

Tuesday -0.16 2.24 30.14 -1.04 4.57 40.27
Wednesday 1.53 0.99 15.53 1.67* 2.14 15.05
Thursday 0.44 0.29 21.66 1.05 0.76 17.44
Friday 3.15*** 1.62 15.29 3.84*** 3.22 13.48
block contracts and the corresponding day-ahead market contracts, i.e. the block
contract market is regarded in this section as a futures market in relation to the
day-ahead market.
The estimated risk premia in block contracts are summarized in table 4.11. Results
are reported for working days only as all results for non-working days are statistically
insignificant. The table contains results for the base load and the peak load contracts.
I apply the Newey-West estimator to receive autocorrelation and heteroscedasticity
robust results.
Table 4.11 reveals the existence of risk premia, in both base load and peak load
contracts, with statistically significant results. A high risk premium, significant
at the 1% level, is observed in base load and peak load contracts, in particular on
Mondays. In addition, significant risk premia in the base load contracts are observed
on Wednesdays and Fridays as they are also in the peak load contracts on Fridays.
Overall, the risk premia in both contracts are significant at the 5% level and have
a magnitude of 0.79 Euro in the base load contracts and of 1.55 Euro in the peak
load contracts. Compared with the average prices of the base load and peak load
contract the results imply that on average 1.66% and 2.54%, respectively, of the
block contract price is paid for the hedging of price risk. For the block contract with
delivery on Mondays the ratio even goes up to around 5%.
155
The obtained results seem to confirm the theory discussed above. Market partic-
ipants are apparently willing to pay a risk premium to secure future prices. The
earlier the hedging is possible, the more market participants are willing to pay.
From this perspective, especially the risk premium in block contracts with delivery
on Mondays should be relatively high since these contracts have a time-to-delivery
of three days, hence forcing market participants to forecast the spot price three days
in advance. The high observed risk premia in this contract match the theoretical
prediction. Furthermore, the observation that the block contracts with delivery on
Mondays were considerably less often traded than the other contracts can also be
regarded as support for the high risk of these contracts.
The obtained results for the block contract market confirm the results of Ronn &
Wimschulte (2009). Although pursuing a different goal, the authors analyze the
sample period August 1, 2002 to September 30, 2007 and estimate the risk premia
in the block contracts. Using an extended sample period the results are consistent
with the results reported in table 2 of their paper.
Risk Premia in the Day-Ahead Market

Risk premia in day-ahead market contracts are estimated as the price difference
between the day-ahead market and the intraday market contracts with the same de-
livery hour. In this analysis the day-ahead market is regarded as a futures market in
relation to the intraday market. I use the last price of the intraday market contracts
and separately estimate the risk premia in contracts with delivery on working and
non-working days.
Tables 4.12 and 4.13 contain the results for the risk premia estimation in day-ahead
market contracts. I report the average risk premium, the median value as a ro-
bustness check and the standard deviation.43 Both the absolute and relative risk
premia are reported. The Newey-West estimator is used to get autocorrelation and
heteroskedasticity consistent results.
The average risk premium on working days amounts to minus 0.10 Euro. For non-
working days the analysis yields an average risk premium of 1.89 Euro. The average
risk premia on non-working days are statistically significant at the 1% level; for the
risk premia on working days no statistical significance is found. For working days the
median is slightly higher, for non-working days slightly lower. However, the mean
and median are still in the same range. It stands out that the daily average risk
premia seem to be higher on non-working days than on working days.
43
The median value is perhaps a better measure to analyze the ex post risk premia (estimated
with a small dataset) since the average value is sensitive to price spikes in one of the markets.
156
Table 4.12
Risk Premia in Day-Ahead Market Contracts (Working Days)
Risk premia in hour contracts traded in the day-ahead market on working days. Both the absolute and relative risk premia are reported. ***, **
and * indicate significance at the 1%, 5% and 10% level; the Newey-West estimator is used to obtain robust standard errors.
Absolute Relative Absolute Relative
Hour Mean Median Std.Dev. Mean Median Std.Dev. Hour Mean Median Std.Dev. Mean Median Std.Dev.
1 1.37** 1.03 14.73 -1.28 3.13 109.55 13 1.01 1.84 15.57 0.65 2.56 23.24
2 0.92 0.84 13.52 -8.79 2.39 180.72 14 0.84 2.08 13.87 0.73 3.32 21.97
3 0.73 1.60 12.53 -6.71 4.58 206.94 15 0.46 1.21 13.51 0.6 2.31 22.51
4 0.82 0.58 14.31 -240.55 2.22 4093.19 16 -0.65 0.10 13.14 -1.92* 0.18 22.70
5 0.43 0.31 14.22 -141.24 1.44 2845.25 17 -1.07* 0.05 14.00 -1.61 0.14 23.08
6 2.21*** 2.66 14.22 7.22* 6.82 85.51 18 -1.61** -0.65 15.91 -2.68*** -1.07 21.91
7 -0.85 -0.07 13.15 -3.27** -0.15 31.83 19 -2.38*** -0.41 21.33 -3.23*** -0.65 28.01
8 0.17 1.59 16.86 -2.22 2.55 31.27 20 -0.53 0.50 15.70 -1.02 0.69 22.61
9 -5.33*** -2.54 19.51 -8.5*** -3.89 31.31 21 2.21*** 2.03 13.09 2.41** 3.74 22.95
10 -1.99*** -0.39 14.05 -3.15*** -0.62 21.16 22 0.51 0.95 10.76 0.17 1.82 22.18
11 0.08 1.01 17.10 -0.68 1.67 22.97 23 -0.74* -0.73 9.68 -1.86** -1.26 21.03
12 1.15 1.90 21.31 0.43 2.14 24.84 24 -0.04 0.07 9.41 -0.71 0.11 25.74
Overall -0.1 0.56 15.02 -17.01** 1.14 1008.60
A possible explanation for the higher risk premia on non-working days is the ob-
servation that electricity buyers primarily participate in the day-ahead market. In
particular, they seem to avoid the intraday market – due to lack of liquidity – on
non-working days.44 This behavior should result in a positive price difference, i.e.
risk premia, between these two market segments. Furthermore, the intraday mar-
ket, as the electricity buyer cover their demand in the day-ahead market, becomes a
buyers market. The frequent occurrence of negative price spikes, as is seen in tables
4.7 and 4.8, backs this argumentation.45
The hourly risk premia are volatile and they frequently change in sign. This cor-
responds to results from other markets, as for example to the results reported by
Longstaff & Wang (2004) for the PJM market. I find significant risk premia in 8 of
the 24 day-ahead market contracts with delivery on working days. The risk premia
are positive in the early morning hours and in the afternoon. The statistically signif-
icant risk premia are mainly negative and found in the morning and evening hours.
On non-working days 10 statistically significant risk premia are detected. Comparing
the median values and standard deviation of the separate hour contracts only the
44
Cf. Weber (2010) for a discussion of potential reasons for the low liquidity in the German
intraday market.
45
Also in discussion with practitioners active in electricity trading this point of view regarding
the intraday market became clear.
157
Table 4.13
Risk Premia in Day-Ahead Market Contracts (Non-Working Days)
Risk premia in hour contracts traded in the day-ahead market on non-working days. Both the absolute and relative risk premia are reported. ***, **
and * indicate significance at the 1%, 5% and 10% level; the Newey-West estimator is used to obtain robust standard errors.
Absolute Relative Absolute Relative
Hour Mean Median Std.Dev. Mean Median Std.Dev. Hour Mean Median Std.Dev. Mean Median Std.Dev.
1 9.06 2.68 92.04 11288.09 7.19 183139 13 2.75*** 3.09 11.50 3.58** 6.30 26.48
2 5.1 1.03 52.90 -63.9 2.84 1248.37 14 2.45*** 1.87 11.59 15.32 4.19 162.34
3 2.4* 1.40 22.59 447.41 4.99 7734.91 15 3.73*** 2.42 19.91 -13.2 6.28 442.71
4 0.95 -0.06 17.03 -168.61 -0.34 2778.67 16 4.44 1.63 47.41 -3.13 4.04 604.68
5 0.12 -0.09 12.93 -529.24 -0.04 9080.70 17 1.21 1.43 18.10 64.78 3.96 1653.55
6 -0.08 -0.57 13.45 23.19 -0.74 2355.31 18 -0.79 -0.34 13.18 -4.49* -0.60 38.46
7 -2.88** -3.12 18.18 459.88 -7.62 7636.55 19 -0.59 0.93 14.83 -1.93 1.88 32.02
8 -1.4 -1.64 16.13 -100.47 -3.23 7861.78 20 0.64 0.69 14.35 1.21 1.51 29.02
9 1.12 0.42 12.23 705.54 1.19 12025.4 21 1.71* 0.97 13.11 2.75 2.46 28.68
10 0.24 0.09 11.11 -0.96 0.40 39.34 22 2.53*** 2.37 14.03 6.17** 4.51 47.05
11 1.73*** 1.24 9.84 2.11 2.67 22.80 23 1.33 1.13 12.55 1.85 2.59 29.28
12 3.26*** 2.97 10.01 4.41*** 5.09 22.65 24 6.42** 2.12 50.61 -6.7 5.48 221.81
Overall 1.89*** 1.01 28.99 490.78 2.88 37066.1
hour 19 contract on working days stands out. The explanation for the high volatility
is one extreme price observation in the day-ahead market.46 The highest positive
risk premia on working days is found in the hour 6 and hour 21 contracts, the lowest
in the hour 9 contract. Using the average prices of the intraday market to estimate
the risk premia in the individual hour contracts, I find no systematic differences in
the results.
Regarding the results for the relative risk premia, the high volatility of the prices in
the two spot market segments is clearly visible in the results. In particular the results
for non-working days, and here in the early morning hours, are strongly influenced
by outliers.47
To illustrate the high volatility of the risk premia the time series of the risk premia
in four selected hour contracts are plotted in figure 4.15.
The selected hour contracts shown in figure 4.15 are equally spread over the day,
with a constant five hour gap in-between. Thus, the risk premia in hour contracts
with fundamental different demand profiles are shown. The vertical axes of the four
46
The extreme observation in hour 19 occurs on November 07, 2006 with a price of 2,436.63
Euro. The exclusion of this price observation from the dataset results in a volatility for the hour 19
contract which is comparable to the volatility of the other hour contracts.
47
I do not adjust the dataset for outliers as the analysis of the absolute risk premia is for the
spot market in the foreground.
158
Figure 4.15
Selected Risk Premia Time Series in Day-Ahead Market Contracts
Time series of risk premia in day-ahead market contracts for selected hours on working days. Sample period: September 25, 2006 to June 30, 2010.
Hour 6 (5-6 am), hour 12 (11-12 am), hour 18 (5-6 pm), and hour 24 (11-12 pm) are displayed. The vertical axis is set between -50 and +50 Euro.
Hour 6 Hour 12
50 50
25 25
Risk Premium [Euro]
Risk Premium [Euro]

0 0
-25 -25
-50 -50
Hour 18 Hour 24
50 50
25 25
Risk Premium [Euro]
Risk Premium [Euro]
0 0
-25 -25
-50 -50
graphs in figure 4.15 are set between minus and plus 50 Euro to amplify the different
volatility. The last price in the corresponding intraday market contract is used to
compute the data. As it can be seen, the risk premia time series for hour 12 and hour
18 show a significantly higher volatility compared to the other two hour contracts.
A form of volatility clustering can be observed in the risk premia.
Time Variation Risk Premia

Another interesting question regarding risk premia in electricity markets is time
variation, i.e. the existence of seasonality and the existence of a trend in the observed
magnitudes. Hadsell & Shawky (2007) report for the New York wholesale market
high risk premia in winter and summer months, according to the yearly demand
pattern. Lucia & Torro (2008) report seasonality in risk premia in week futures
traded at the Nord Pool. For the analysis of the data regarding the existence of
seasonality in the risk premia, I have the choice to either analyze all hour contracts
separately or to use the daily average or some blocks of hours. The estimation of
the daily risk premia – calculated as the daily price in the day-ahead market minus
the daily price in the intraday market – and the estimation of the average monthly
risk premia based on the daily results seem to be the most straightforward approach.
159
Table 4.14
Risk Premia in Day-Ahead Market Contracts by Delivery Month (Working Days)
Risk premia in day-ahead market contracts by delivery month. The estimated risk premia are shown for three blocks of hours on working
days. ***, ** and * indicate significance at the 1%, 5% and 10% level; the Newey-West estimator is used to obtain robust standard errors.
Base Peak Off-Peak
Mean Median Std.Dev. Mean Median Std.Dev. Mean Median Std.Dev.
January -2.49*** -0.85 15.49 -4.3*** -1.82 18.14 -0.61 0.00 11.66
February 0.01 0.71 14.26 -0.82 0.54 16.17 0.73 0.24 10.96
March 0.15 0.02 11.72 -0.01 0.14 10.93 0.32 -0.19 12.54
April 0.63 0.99 13.11 0.83 1.59 15.36 0.42 0.42 9.95
May -2.6*** -0.04 12.49 -2.25*** 0.04 13.21 -2.95*** -0.43 11.62
June 0.06 0.24 14.70 -0.07 0.55 17.78 0.2 0.00 10.24
July 1.71*** 2.88 14.03 -0.1 2.40 17.14 3.56*** 3.48 9.27
August 0.44 1.03 11.14 -1.04* 0.50 11.92 1.96*** 1.62 10.10
September 1.32*** 0.97 13.89 0.57 0.51 16.34 2.08*** 1.53 10.56
October -0.31 -1.25 19.73 -0.14 -1.89 23.01 -0.5 -0.84 15.09
November 0.09 0.08 60.95 -3*** -0.05 81.22 3.24*** 0.82 19.69
December 1.1'** 0.99 20.83 0.92 0.89 25.82 1.29** 1.11 12.44
However, this procedure is not practicable due to the significant number of hours
without trading in the intraday market.48 I decide to evaluate the existence of
seasonality by applying the following procedure: first, I estimate the risk premia in
all hour contracts. Second, I reorder the estimated risk premia by months. Third, I
calculate the average risk premia for three blocks of hours: base (0 – 24 am), peak (8
am – 8 pm) and off- peak (0 am – 8 am and 8 pm – 0 am); the choice of this blocks
is based on the results obtained above which show that these hour blocks exhibit
similar characteristics. Table 4.14 and table 4.15 summarize the results.
Significant positive risk premia on working days are found for summer months; re-
garding the base block, contracts with delivery in July and September exhibit statis-
tically significant risk premia. Significant negative risk premia are found for January
and May. The results for the other months are mixed and not significant. Carefully
interpreting the results and taking into account the short sample, the existence of
a seasonality in the risk premia in the spot market cannot be detected. This is
confirmed by the results for non-working days in table 4.15. In contrary, Viehmann
(2009) analyzes four selected hours and finds – using data from the day-ahead market
and OTC prices – significantly higher risk premia in winter months.
Finally, I pose the question whether the risk premia changed over the last years. The
question is whether the sign or magnitude of the risk premia are constant or evolve
48
Off-peak hours have the highest number of days without trading. It may be hence assumed
that the daily average price is upward biased.
160
Table 4.15
Risk Premia in Day-Ahead Market Contracts by Delivery Month (Non-Working Days)
Risk premia in day-ahead market contracts by delivery month. The estimated risk premia are shown for three blocks of hours on non-working
days. ***, ** and * indicate significance at the 1%, 5% and 10% level; the Newey-West estimator is used to obtain robust standard errors.
Base Peak Off-Peak
Mean Median Std.Dev. Mean Median Std.Dev. Mean Median Std.Dev.
January -1.38*** -0.53 12.97 -0.65 0.29 13.52 -2.12*** -1.42 12.35
February 0.45 1.46 12.71 -1.2 0.08 12.79 2.15*** 2.58 12.41
March 0.84* 0.74 12.34 0.32 0.34 10.47 1.38* 1.16 14.02
April 2.04*** 2.11 10.92 2.85*** 3.16 10.49 1.2 0.38 11.30
May 1.1*** 0.66 10.63 2.16*** 1.90 9.27 0.01 -0.44 11.77
June 0.15 0.65 10.51 0.99* 1.11 9.25 -0.7 0.30 11.60
July 0.1 0.07 9.20 -0.15 -0.46 8.01 0.36 0.53 10.30
August 3.8*** 3.09 12.67 4.71*** 4.26 11.57 2.86*** 2.07 13.69
September 2.38*** 1.04 11.95 3.77*** 2.09 11.49 0.94 -0.99 12.27
October 13.95*** 3.13 103.29 8.23** 3.30 59.51 19.87** 3.11 134.25
November -1.77** -1.28 13.34 -3.36*** -1.95 13.70 -0.16 -0.33 12.79
December 4.94*** 2.93 21.85 4.19*** 3.12 17.50 5.75*** 2.74 25.75
over time. To answer this question I decide to analyze the daily risk premia in the
day-ahead market. These premia are calculated as the average of the 24 hourly risk
premia. In the case that no trading in one of the hour contracts in the intraday
market occurred, the daily risk premia is calculated with less than 24 values. The
resulting time series is displayed in figure 4.16.
The y-axis in figure 4.16 is set between minus 60 and 60 Euro. The application
of various simple methods to estimate whether a time evolution of the risk premia
takes place yields no clear results. This is probably due to the high volatility at
the end of the sample period after a strong increase over the last two years. Further
research on the evolution of the risk premia is necessary49 , given an extended dataset
is available.50
49
When testing for the evolution of the risk premia throughout the analyzed sample period, I find
no obvious trend. From beginning on the risk premia seem to be positive and extremely volatile.
The negative daily risk premia obtained by Daskalakis & Markellos (2009) for the first year of the
intraday market’s existence are probably due to calculation of daily prices on the intraday market
as arithmetic averages of the hourly prices and the high number of non-trading hours in this period.
50
Similar to Boogert & Dupont (2005) I examine the practical relevance of the hourly risk premia
by testing two simple (spread) trading strategies over the period August 2008 to May 2009. A spread
strategy in electricity markets consists of a long position in one market segment and a short position
in the other one. Therefore, I examine the two hours with the largest positive and negative risk
premia over the whole sample period (on working days), hour 9 and hour 12. For the hour with
a positive risk premium the strategy to be tested is a short position in the day-ahead and a long
position in the intraday market. For the hour with a negative risk premium the opposite strategy
applies. The profit for the first strategy (per MWh) is the day-ahead price minus the intraday price,
for the second strategy the intraday minus the day-ahead price. I start the test in August 2008,
because from this month on there is no occurrence of non-trading days in the hours of interest.
161
Figure 4.16
Time Variation Daily Risk Premia Day-Ahead Market
Relative risk premia in the day-ahead market on a daily scale. Sampel period: January 2, 2008 to June 30, 2010.
The graph only includes working days.
60
20
Risk Premia [%]
Jan. 08 Jul. 08 Jan. 09 Jul. 09 Jan. 10
-20
-60
Source: Own work.
Term Structure of Risk Premia

The time period between September 25, 2006 and August 30, 2008 offers the unique
opportunity to analyze the German electricity spot market in respect to the existence
of a term structure of risk premia. That is due to the fact that during this time period
the three market segments were simultaneously in existence. Thus, it was possible
to buy and sell electricity for the same delivery period in three different market
segments with the only difference being the trading point-in-time or rather the time-
to-delivery of the specific contract. However, it was not possible to trade electricity
contracts with delivery in every individual hour. Rather the tradable delivery periods
were determined by the block contract market. The tradable delivery periods were
hence the whole day (base load contract), the peak hours (peak load contract), and
Without further investigating the traded volume (and by ignoring transaction costs) I assume that
it is possible to trade an additional volume of 10% at the quoted price in the intraday market. As
results I get a three-digit average profit for hour 9 and a three-digit average loss for hour 12 as well as
a high volatility. Based on these results I wonder whether professional market participants with no
interest in the physical delivery of electricity (speculators, arbitrageurs, etc.) are seriously interested
in investing time and money in trading strategies with such profit-loss potentials. Discussing this
point with representatives of a leading investment bank in Western Europe I received the argument
that based on the low liquidity in the intraday market and on the high volatility of the risk premia
the potential profits of an arbitrage strategy based on the risk premia are by far not sufficient to
justify an engagement.
162
the off-peak hours (through a synthetic contract: a long position in the base load
contract and a short position in the peak load contract).
For the analysis of the term structure, I relax the restriction to the intraday market
data. Otherwise, the sample period with overlapping trading in the three market
segments would be only eight months. However, the obtained results have to be inter-
preted with caution as the liquidity during parts of the sample period was extremely
low.
The question whether market participants are willing to pay different risk premia
depending on the time-to-delivery is of high importance, for both theoretical and
empirical purposes. The results above indicate that similar to other markets the
German spot market is characterized by positive risk premia. However, based on
the above results no conclusion concerning a term structure of risk premia can be
drawn. Empirical results on the existence of a term structure of risk premia in fu-
tures markets are mixed. Shawky et al. (2003) find that the risk premia in futures
with delivery at the California-Oregon Border, traded at the New York Mercantile
Exchange, are an increasing function of time-to-delivery. Weron (2008) and Marck-
hoff (2009) on the other side offer empirical evidence for a decreasing risk premia
with increasing time-to-delivery. From a theoretical point of view, the framework de-
veloped by Benth et al. (2008) is able to explain a term structure of risk premia with
changing risk preferences and hedging demand across different maturities. All em-
pirical results obtained to date deal with maturities in the range of weeks or months.
To my best knowledge, I am the first to have the possibility to research the term
structure of risk premia on such a short time scale. The already obtained results
lead to the expectation to find higher risk premia in the block contract market.
When analyzing the available data the low liquidity in the block contract and in the
intraday market has to be considered. In particular at the beginning of the sample
period defined above, when trading in the intraday market had just been introduced
and at the end, when trading in the block contract market was coming to an end.
For this reason, I reorder the dataset. For the analysis of a specific contract only
days with trading in all hours of interest in all three market segments can be included
in the final dataset. Originally, the sample period extends to over 706 trading days,
486 of them being working days.
I begin to reorder the data with the working days. The reduction of the dataset
to days when trading in all hours of interest in the intraday market took place,
results in 174 working days with trading in all hours and 350 working days with
trading in the peak hours. In a second step, I sort out all days without trading in
the corresponding block contract. I get 147 working days for the base hours and
285 working days for the peak hours at which at least one trade in all three market
163
Table 4.16
Term strucutre of risk premia. The risk premia are estimated with the last price on the intraday market on working days. ***, ** and *
indicate significance at the 1%, 5% and 10% level; the Newey-West estimator is used in order to obtain robust standard errors.
Block Contracts Day-Ahead Contracts
Mean Median Std.Dev. Mean Median Std.Dev.
Base 1.60 1.62 11.63 0.42 1.16 9.47
Peak 3,04*** 1.92 18.10 1,76** 1.04 15.17
Off-Peak 0.07 0.55 6.73 0.33 0.37 5.36
segments and hours of interest took place. For the off-peak hour contracts trading
in all off-peak hours in the intraday market and in both the peak and the block
contract in the block contracts market is necessary. The reordering shows that this
has been the case on 132 working days. Conducting the same reordering procedure
for the non-working days results in an extremely small dataset. Therefore, I decide
to forgo the non-working days. Thus, the following analysis only deals with working
days.
After the reordering of the data, I estimate the risk premia at two points-of-time, one
being the trading in the block contract market and the other one in the day-ahead
market. This results in the estimation of two risk premia in contracts with identical
delivery periods and different time-to-deliveries. The results allow me to evaluate
whether a term structure of risk premia on such a short time scale is apparent. The
results for the two markets and three contracts are shown in table 4.16.
Due to the skewness of the distributions of the risk premia table 4.16 contains the
average and the median risk premia. I again employ the last price from the intraday
market. The risk premia in the block and in the day-ahead market contracts are
estimated as the price difference between the particular market and the intraday
market. That is a significant difference compared to the section above, where the
risk premia in the block contracts were estimated as the price difference between
the block contract and the day-ahead market. The results for the base and off-peak
hours are insignificant although in the off-peak hours the risk premia seem to be
higher in the day-ahead than in the block contracts. The risk premia are higher in
the block contracts for the base hours than in the day-ahead contracts. For the peak
contracts – as mentioned the most liquid ones – I get statistically significant results.
The risk premium in the peak load contract is on average 3.04 Euro and significant
164
at the 1% level. For the day-ahead peak hours I get a risk premium of 1.76 Euro,
significant at the 5% level.
Market participants were apparently willing to pay a higher risk premium for the
possibility of an earlier hedge during the period when all three market segments of
the spot market were active. For the peak hours, for example, the risk premia in
the block contracts are around 70% higher than for the day-ahead market contracts.
This results are in accordance with the findings above.
Drivers of Risk Premia

When investigating potential drivers of the risk premia the equilibrium model of
Bessembinder & Lemmon (2002) provides a relation between the anticipated distri-
bution of the expected spot price and the ex ante risk premium. It identifies the
third and fourth moment of the price distributions as determinants of the risk pre-
mia. With the methodology proposed by Longstaff & Wang (2004), this theoretical
model can be transformed into an empirically testable relation. The ex post risk pre-
mia πi (T ) are regressed on the variance, V ARi [S(T )], and skewness, SKEWi [S(T )],
of the corresponding spot prices in this analysis. The skewness in this case is non-
standardized. The relation is defined as
πi (T ) = a + b · V ARi [S(T )] + c · SKEWi [S(T )]. (4.8)
Bessembinder & Lemmon (2002) show that the relation between the risk premia and
the variance – under certain conditions – is negative and between the risk premia
and the skewness positive.
I regress the 24 hourly ex post risk premia in day-ahead market contracts on the
variance and skewness of the corresponding price distributions, for both working
and non-working days, and find no significant results for the coefficients. Thus, I do
not report the results.
These findings are contrary to Longstaff & Wang (2004) who find a significant relation
for the PJM day-ahead market. To my best knowledge Ronn & Wimschulte (2009)
are the only ones testing the relation for the German spot market. They use the risk
premia in futures traded at the Austrian exchange with delivery in Germany and the
prices in the day-ahead market of the EEX. The authors distinguish also between
working and non-working days. The results for working days are all insignificant, for
non-working days only the relation for the risk premia on Sundays are found to be
significant.
165
Table 4.17
Risk Premia in Month Base Futures
Risk premia in month base futures. The risk premia are calculated with monthly data. ***, ** and * indicate significance at the 1%, 5% and
10% level; the Newey-West estimator was used in order to obtain robust standard errors.
Maturity Mean t-value Std.Dev. Mean t-value Std.Dev.
One 2.16** 2.11 8.53 3.42 1.59 18.23

Two 2.83* 1.78 11.10 3.57 1.15 22.70
Three 3.15 1.57 12.34 3.67 0.97 23.97
Four 3.15 1.35 13.45 3.22 0.75 25.17
Five 3.2 1.22 14.68 2.58 0.55 26.71
Six 3.09 1.10 15.53 1.74 0.34 28.48
4.4.3.2 Risk Premia in Futures Contracts
The results for the futures market are reported in four parts. First, the estimated
risk premia are reported. Afterwards, the potential term structure of risk premia is
discussed and then the time-dependence is analyzed. Finally, potential drivers are
addressed.
Risk Premia in the Futures Market

Using the monthly spot prices which were shown in figure 4.10, I estimate the risk
premia in the month futures. The estimation follows equation (4.2) for the absolute
risk premium and equation (4.7) for the relative risk premium. The futures prices are
aggregated to monthly prices to overcome autocorrelation problems. The aggregation
of the data results in a shortening of the time series for every futures contract from
approximately 150 observations to six monthly prices. Every monthly price is used
for the computation of the risk premium with corresponding time-to-delivery.51
The results are reported in tables 4.17 and 4.18 for the base and peak futures,
respectively. Standard errors are calculated autocorrelation and heteroscedasticity
robust using the Newly-West estimator. The standard deviation and the t-value are
reported as well.
The absolute risk premia exhibit a similar evolution for both the base and peak
futures. A steadily increase with the time-to-delivery is observed. When the risk
premium in the one month base futures accounts to 2.16 Euro, the risk premium in
the six month futures is with 3.09 Euro approximately 45% higher. For the peak
51
All estimations are also performed on the daily data. The results are similar to the results on
monthly data except lower standard errors due to the autocorrelation.
166
Table 4.18
Risk Premia in Month Peak Futures
Risk premia in month base futures. The risk premia are calculated with monthly data. ***, ** and * indicate significance at the 1%, 5% and
10% level; the Newey-West estimator was used in order to obtain robust standard errors.
Maturity Mean t-value Std.Dev. Mean t-value Std.Dev.
One 4.06** 2.29 15.70 4.81* 1.85 23.26

Two 5.62** 2.13 19.71 5.57 1.54 28.13
Three 6.16'* 1.93 20.96 5.97 1.41 28.78
Four 6.24* 1.74 21.80 6 1.29 28.96
Five 6.54 1.66 22.95 5.99 1.18 29.91
Six 6.54 1.54 24.33 5.31 0.95 6.00
futures the risk premium in the one month futures accounts to 4.06 Euro and for
the six month futures to 6.54 Euro; the increase is approximately 60%. However,
a look at the relative risk premia questions this monotone increase. Here the risk
premia in the base futures reache a maximum in the third month futures and decrease
afterwards. In the peak futures the maximum is observed in the four month futures.
Regarding the statistical reliability of the results, the risk premia in the one and two
month base futures are significant. For the peak futures the risk premia in the one to
four month futures are found to be significant. The significance of the relative risk
premia is weak; only the risk premium in the one month peak futures is significant.
The obtained significant results for the risk premia confirm the hypothesis that elec-
tricity consumers seem to mainly use short-term futures for hedging purposes. The
evidence for decreasing risk premia in futures with longer time-to-delivery is in accor-
dance with the theory as well; a decreasing demand of the electricity consumer could
results in a decrease of the hedging pressure together with the risk premia. Thus, a
term structure of risk premia could exist in the German market in accordance with
the theoretical predictions with the model of Benth et al. (2008).
The estimated relative risk premium accounts for around 3% of the price of the one
month base futures and for around 5% of the month peak futures. Compared to other
futures markets, this is a relatively large risk premium which the market participants
are seemingly willing to pay for the disposal of price risk for a time horizon of one
month.
To analyze the time evolution of the risk premia, figure 4.17 shows the relative risk
premium in the one month base and peak futures over the sample period.
Figure 4.17 reveals that the relative risk premia are highly volatile and change regu-
167
Figure 4.17
Relative Risk Premia in One Month Futures
Relative risk premia in the one month future with respect to the futures price. The straight line represents the month
base future, the dashed line the month peak future.
60
30
0
Jul 02 Jul 03 Jul 04 Jul 05 Jul 06 Jul 07 Jul 08 Jul 09
Risk Premium [%]
-30
-60
-90
-120
larly in sign. However, tt can be assumed that large parts of these two effects are due
to forecast errors. Theses errors result in partially dramatic discrepancies between
futures and realized spot prices, in parts exceeding 50%. These discrepancies are by
far greater than the estimated average risk premia which are in the range of 5%. Re-
garding the change in sign no systematic patterns are visible. The question whether
the risk premia – both in sign and magnitude – evolve over time is hence difficult
to answer. Further research needs to be done as a longer sample period is available
since the evolution of risk premia over time is of high interest. The hypothesis is that
the market entry of new market participants together with a learning curve should –
at least from a theoretical point of view – lead to a more efficient market and hence
to a decrease of the risk premia.

The results in tables 4.17 and 4.18 provide evidence for the existence of a term
structure of risk premia in the futures markets. I decide to analyze the daily data
with the aim to find further support for a potential term structure. Therefore, I
compute the absolute risk premia for every daily observation and synchronize the
calculated risk premia according to the first day of the delivery month. This allows
168
Figure 4.18
Risk Premia in Month Base Futures by Time-to-Delivery
Risk premia in the month base futures with respect to time-to-delivery. The futures are synchronized according to the
delivery month. In addition for graphical reasons a moving average over seven days is also shown (straight line).
6
Risk Premium [Euro]
0
1 31 61 91 121 151
Days-to-Delivery
to sort all the obtained risk premia according to their time-to-delivery. The results
of this computation are found in figure 4.18 for the six first month base futures.
The straight line in figure 4.18 is a moving average over seven days to smooth the
daily risk premia data. This average is, in accordance with the results for the monthly
data obtained above, increasing. An interpretation of this increase as a term struc-
ture of risk premia, i.e. increasing risk premia with increasing time-to-delivery may
be postulated. However, absolute risk premia exhibit the problem that changes on
the price level may bias the results. Thus, the use of the relative risk premia is prob-
ably more meaningful in this case. The results of the same computation as above for
the relative risk premia are found in figures 4.19 and 4.20. The relative risk premia
in the month base and month preak futures, respectively, and in dependence of the
time-to-delivery are depicted in these figures.
In both figures every data point is calculated, similar to the absolute risk premia in
figure 4.18, on average from 60 separate observations. For a better vizualisation also
here a moving average over seven days is displayed.
The results for the relative risk premia are different to the of the absolute risk
premia. The relative risk premia in the base month futures, for example, indicate
a term structure with an increasing or stable risk premium at the short-end and
169
Figure 4.19
Relative Risk Premia in Month Base Futures by Time-to-Delivery
Relative risk premia in the month base futures with respect to time-to-delivery. The futures are synchronized according to
the delivery month. In addition for graphical reasons a moving average over seven days is also shown (straight line).
6
Risk Premium [%]
0
1 31 61 91 121 151
Days-to-Delivery
Source: Own work.
Figure 4.20
Relative Risk Premia in Month Peak Futures by Time-to-Delivery
Risk premia in the month peak futures with respect to time-to-delivery. The futures are synchronized according to the
delivery month. In addition for graphical reasons a moving average over seven days is also shown (straight line).
10
8
Risk Premium [%]
0
1 31 61 91 121 151
Days-to-Delivery
Source: Own work.

170
then a decreasing risk premium with increasing time-to-delivery. Unfortunately, the

dataset does not include long-term futures to test whether this trend is also observed
for futures with longer time-to-delivery. The results for the peak futures are rather
pointing to a flat term structure. However, a high volatility at the long-term is
visible. These results are in accordance with the results for the spot market which
were discussed in the last section. I find for the spot market an increasing risk premia
at the short-end. When comparing this with figure 4.19 also in the futures market
an increase at the short-end is found.
Shawky et al. (2003) were the first to report results on the relationship between risk
premia and time-to-delivery. For the years 1998 and 1999 they find a linear increasing
risk premium with increasing time-to-delivery for the California-Oregon Border area.
On the other side, Diko et al. (2006) find positive short-term and negative long-term
risk premia in OTC forward prices for three European futures markets. Decreasing
risk premia with increasing time-to-delivery are reported by Marckhoff (2009) for
the Nord Pool market. Weron (2008) finds the same effect by modeling the market
price of risk for the Scandinavian area through stochastic models. Benth et al. (2008)
develop a theoretical model to explain this effect.
Posing the question whether the risk premia is stable over the sample period, I
calculate the term structure for every full year in the sample period. Therefore, I
divide the data which were used for the estimation of the term structure above by
years. I use the month base futures. A term structure for 2007, for example, is
estimated with all price observations of the first six month futures in this year. The
futures are again synchronized to the delivery month. The results of this procedure
for the years 2003 to 2009 are shown in figure 4.21.
The results of the estimation on a yearly scale show that the term structure is not
stable. In 2005, for example, almost perfect linearly decreasing risk premia with
time-to-delivery are observed. In opposite, in 2009 almost perfectly increasing risk
premia are found. In both cases the relative risk premia at the long-end of the term
strucutre account to around 40% and -40%, respectively. The other years are between
these two extreme cases. Except for 2008, where also significantly decreasing risk
premia with increasing time-to-delivery are observed, the other yearly estimations
reveal either an almost monotone term structure or an increase at the short-end and
a decrease afterwards.
The results displayed in figure 4.21 pose the question on the stability of the aggre-
gated risk premia in figures 4.19 and 4.20. Further research is necessary on this
topic, but it seems that, based on the obtained results, a (stable) term structure of
risk premia in the futures market has to be denied. Regarding the short-end, i.e. the
one and two month futures, it seems possible to conclude that a positive risk premia
171
Figure 4.21
Relative Risk Premia in Month Base Futures by Year
Relative risk premia in month base futures by time-to-delivery and by year. All price observations for the six month
futures in a year a used for the calculation of the term strucutre in this year.
60
40
20
Risk Premia [%]
0
1 31 61 91 121 151
-20
-40
2003 2004 2005 2006

2007 2008 2009
-60
Source: Own work.
is observed over the years. Regarding the long-end, on the other side, no conclusion
can be drawn as the results are too mixed. However, it seems that for most of the
time decreasing risk premia occur at the long-end.
Regarding future research, it is of interest whether it is possible to identify factors
which determine the change in the risk premia at the long-term over the years. For
example the possibility exists that trading strategies based on recently observed risk
premia are responsible for the regular changes as market participants try to profit
from the observed price differences.
Time Variation Risk Premia

The time variation in the occurrence of risk premia gained significant research in the
last years. Lucia & Torro (2008), for example, find seasonality in the risk premia at
the Nord Pool. Their results indicate that risk premia are highest and statistically
significant for delivery periods in winter and zero for delivery periods in summer.
Cartea & Villaplana (2008) model the magnitude and sign of risk premia depending
on demand and capacity.52 One implication of their model is the occurrence of posi-
tive risk premia which are caused by high volatility of demand. This implies positive
52
Cf. section 2.3.3.2 for a discussion of the model developed by Cartea & Villaplana (2008).
172
Table 4.19
Risk Premia in Month Base Futures by Delivery Period
Average risk premia in the analyzed individual month base future contracts which are observed over their whole trading period. ***, ** and * indicates
significance at the 1%, 5% and 10% level; the Newey-West estimator was used in order to obtain robust standard errors.
FUTURE 2003 2004 2005 2006 2007 2008 2009 2010 AVERAGE
January -0.8 8.34 6.85 -12.27 35.84 10.37 26.93 6,23 10.19
February -6.74 9.49 -2.76 -12.7 32.99 8.02 30.72 6,91 8.24
March -2.15 0.16 -11.09 -7.19 27.47 5.83 25.68 3,75 5.31
April 0.06 3.7 -7.65 7.25 9.47 -10.71 16.91 -0,49 2.32
May 0.42 0.28 -6.03 13.67 1.71 -2.58 10.04 -4,29 1.65
June -6.45 2.48 -11.81 11.05 4.58 -12.71 8.05 -2,16 -0.87
July -12.86 1.71 -8.33 -22.26 15.78 -1.31 6.05 - -3.03
August -7.7 -0.55 2.14 7.1 10.93 2.93 -1.71 - 1.87
September -0.36 -0.5 -4.39 8.76 5.45 -15.24 -0.52 - -0.97
October -4.72 3.21 -2.13 11.11 -16.14 -6.97 -2.11 - -2.54
November 1.72 6.32 -21.49 13.54 -14.02 25.73 13.23 - 3.58
December 3.7 5.09 -12.98 23.72 -0.93 28.06 10.21 - 8.13
AVERAGE -2.99 3.31 -6.64 3.48 9.43 2.62 11.96 1.66 -
risk premia in winter months for the most markets. The model of Bessembinder &
Lemmon (2002) also suggests the existence of seasonality in risk premia is caused by
demand uncertainty.
Due to the lack of a sufficient sample period, I cannot directly test for seasonality in
the German futures market. I only have at maximum eight futures with delivery in
a particular month, meaning that only eight independent expectation building pro-
cesses regarding a particular calendar month took place. To overcome this problem,
I calculate the average risk premia contained in every individual futures contract.
However, I do not calculate the significance of these individual risk premia as the
price time series of an individual futures is highly autocorrelated. Thus, the price
difference between the futures price time series and the spot price in the delivery
month is almost always highly significant. However, this significance is probably in
large parts due to forecast errors rather than to the systematic occurrence of risk
premia. This point of view is supported by the change in sign of the risk premium
which cannot be explained solely based on risk consideration arguments.53 Price
shocks with asymmetrical impact on spot and futures prices are one explanation for
these forecast errors. The results for the base futures are reported in table 4.19.
As expected the calculated risk premia exhibit a high variability both in magnitude
and sign. Therefore, I report the average for every particular month in the last
53
Cf. Redl et al. (2009), p. 362.
173
Table 4.20
Risk Premia in Month Peak Futures by Delivery Period
Average risk premia in the analyzed individual month peak future contracts which are observed over their whole trading period. ***, ** and * indicates
significance at the 1%, 5% and 10% level; the Newey-West estimator was used in order to obtain robust standard errors.
FUTURE 2003 2004 2005 2006 2007 2008 2009 2010 AVERAGE
January -2.69 20.58 13.35 -23.59 53.14 17.77 41.76 13.59 16.74
February -7.14 22.76 -1.19 -15.49 53.47 18.78 49.14 12.88 16.65
March 0 5.01 -14.6 -6.19 47.32 10.16 42.65 8.92 11.66
April 3.8 10.83 -6.88 12.71 15.19 -15.56 25.85 3.49 6.18
May 1.62 1.96 -4.72 17.94 1.62 -5.09 13.66 -5.6 2.67
June -13.03 5.02 -19.51 17.79 3.21 -18.43 14.1 -2.13 -1.62
July -21.85 5.84 -8.55 -61.18 31.79 6.04 12.14 - -5.11
August -12.08 4.54 5.59 14.76 26.2 12.13 -2.67 - 6.93
September 4.66 3.56 -4.27 12.9 14.95 -14.95 -1.31 - 2.22
October -4.09 6.66 -1.67 18.14 -21.04 -10.11 -6.5 - -2.66
November 9.35 11.22 -43.62 15.1 -30.14 35.48 19.7 - 2.44
December 10.36 10.18 -25.84 36.69 -3.57 40.69 14.9 - 11.92
AVERAGE -2.59 9.01 -9.33 3.30 16.01 6.41 18.62 5.19 -
column. I find evidence for seasonality when comparing the averages. Positive risk
premia are observed in winter months, i.e. from December to February. After a
decrease in spring and autumn the summer months seem to be characterized by
negative risk premia. However, no statistical verification of these results is possible
as the averages are calculated over a maximum of eight observations. The results
are confirmed by similar results for the peak futures reported in table 4.20.
The obtained results seem to be in accordance with the empirical literature on elec-
tricity futures market. They confirm that high risk premia tend to occur in winter
months, in the German market in particular in the period December to February.
On the other side, the risk premia in the summer months are significantly lower and
oscillate around zero.
Drivers of Risk Premia

The results in the previous section provide evidence for the existence of risk premia
in the German electricity futures market. This section is dedicated to the discussion
of potential drivers. The analysis focuses on the question whether the existence of
risk premia can be linked to risk considerations. Possible fundamental drivers are
discussed qualitatively. The quantitative verification is left for further research.
I test the adequacy of the Bessembinder & Lemmon (2002) model as a potential
174
Table 4.21
Regression Risk Premia on Variance and Skewness of Underlying Spot Price
Regression of risk premia on variance and skewness of spot prices in the delivery month. The skewness is unnormalized. ***, ** and
* indicate significance at the 1%, 5% and 10% level; the Newey-West estimator was used in order to obtain robust standard errors.
Future a b c Adj. R² [%]
Base 2.62*** -0.005509 -0.000024 21.75

Peak 10.34*** -0.002632 -0.000006 19.39
explanation of the observed risk premia. Therefore, I regress according to equation

(4.8) both the month base and the month peak futures prices on the third and the
fourth moment of the spot price distributions in the delivery month. Performing this
regression with monthly averages of the futures prices results in residuals with strong
serial correlation. This is also observed by Redl et al. (2009); the authors hence use
instead the futures price on the last trading day.54 I decide to run the regression in
the same way as Redl et al. (2009) and observe that the serial correlation problem
is solved trough the replacement of the futures prices. However, the analysis is only
performed for the one month futures as in the case of futures with a longer time-
to-delivery the serial correlation is still present. The results of the regression are
reported in table 4.21.
The coefficients for the variance and skewness reported in table 4.21 are not signifi-
cant. Thus, the results do not support the Bessembinder & Lemmon (2002) model;
the assumption that risk premia in the German electricity futures market are linked
to risk considerations is not supported by this analysis.
Redl et al. (2009) analyze the one month futures in the period November 2003 to
May 2008 and find mixed results. Regarding the futures markets in other countries,
Furio & Lucia (2009), for example, analyze the Spanish futures market and find
supporting evidence. Lucia & Torro (2008) report mixed results for the Nord Pool
futures where the dependence holds before a shock period and thereafter vanishes.
Marckhoff (2009) finds support for the model using data from the Nord Pool futures
market. The mixed results reported in the literature suggest that other drivers may
be relevant as well.
Fundamental factors can also serve as drivers of the risk premia. Only a few results
have been reported to date. Douglas & Popova (2008), for example, link risk premia
54
However, the methodology of Redl et al. (2009) is still different as the authors use the variance
and skewness of the spot price in the trading month for their analysis.
175
and gas storage inventories. The authors develop a model which links increasing
gas storage inventories under realistic assumptions to a decrease of the risk premia.
That is explained by a decreasing probability for the occurrence of price spikes.
Daskalakis & Markellos (2009) link risk premia and emission allowance spot prices.
They empirically demonstrate a positive relationship, among others with data from
the EEX.
An empirical analysis of the German electricity wholesale market is conducted in this

chapter. I aim to test the suitability of the risk premia approach as a price formation
mechanism. Therefore, I conduct an in-depth analysis of the German market to
detect the existence of potential risk premia. I apply the ex post approach. The
analysis is divided in two parts: in the first part the focus is on the spot market,
while in the second part I focus on the futures market.
In the first part, I analyze all three market segments of the spot market which are
or were in existence during the sample period which extends between August 2002
and June 2010. These three market segments are the day-ahead market, the block
contract market, and the intraday market. Trading in the day-ahead market in its
current form started in August 2002, while trading in the intraday market started
in September 2006. The block contract market ceased its activity in August 2008.
The analysis of the spot market yields the following results: I find positive risk premia
in the block contract and in the day-ahead market. Risk premia in block contracts
are particularly significant and high for contracts with delivery on Mondays. These
contracts were traded on Friday and hence had a time-to-delivery of three days.
Risk premia in day-ahead market contracts are extremely volatile and change in
sign throughout the day. The average daily risk premium for non-working days is
significant. Furthermore, I detect a term structure of risk premia during the sample
period when all three market segments were active. Risk premia seem to be higher
in contracts with a longer time-to-delivery. In addition, I examine potential time
variation in the risk premia but find no significant results. Also the results for a
potential seasonality are mixed. When testing for a relation between the variance
and skewness of the underlying spot price and the risk premia, a relation proposed
by Bessembinder & Lemmon (2002), I find no significant results.
In the second part, I conduct an analysis of the futures market. Because of liquidity
considerations, I restrict the analysis to month futures with financial settlement. The
sample period extends from July 2002 to June 2010.
176
The analysis of the futures market yields the following results: I find evidence for
positive risk premia in short-term futures, i.e. futures with a time-to-delivery up
to around three months, which are statistically significant. I also detect evidence
for seasonality in the risk premia. The risk premia seem to be positive for delivery
months in winter and zero or even negative in summer. When testing the existence
of a term structure the results are mixed. It seems that in certain periods the
risk premia increase with increasing time-to-delivery, wheras in other periods they
decrease. However, the evidence indicates that the aggregate risk premia decrease in
the long-term. The results for a verification of the Bessembinder & Lemmon (2002)
model are negative.
The obtained results are consistent with both theoretical and empirical literature.
They support the hypothesis that hedging pressure is an appropriate approach for
price formation in the German electricity futures market. The short-term futures
seem to be mainly used by electricity consumers for hedging purposes. With increas-
ing time-to-delivery the demand of electricity consumers seems to decrease. This
results in low and statistically insignificant risk premia in the mid-term and long-
term futures. The question whether the risk premia change sign and thus whether
a market segmentation is apparent cannot be answered due to the shortness of the
sample period.
Further research on this topic seems to be promising and necessary. Regarding
the analysis of the spot market an identification of potential drivers of the risk
premia in the German market would extend the understanding of the price formation
mechanism. In addition – as soon as a larger dataset is available – the time-variation
of the risk premia should be analyzed. The question whether a convergence of the
day-ahead and the intraday prices will take place or whether the risk premia will
persist is of particular interest. Both research avenues are related to the question
whether the observed positive risk premia are an appropriate compensation for the
associated risk or rather an indication of market inefficiency. The liquidity of the
intraday market needs to be further investigated in order to test the robustness
of the results. An interesting research question is whether a relation between the
magnitude of the risk premia and the liquidity exists. Regarding the analysis of
the futures market, future research in at least two directions seems to be promising.
First, the time-evolution of the risk premia in this market segment is of even higher
interest than that in the spot market. As the main part of the liquidity in the German
electricity market is found in this market segment, it should also be the market with
the highest efficiency. To extend the used dataset the quarter and year futures could
also be included when a sufficient liquidity and length of the time series is reached.
Furthermore, when investigating the time evolution of the magnitude and sign of the
177
risk premia, the question on a constant average risk aversion could be posed. Second,
an identification of fundamental drivers for the risk premia seems to be promising.
The role of fuels (coal, gas, and oil) and of emission allowances is here of particular
interest. As a relation is found for other markets the question is whether similar
mechanisms are valid in the German market. However, other determinants – due to
the different generation technologies, certain determinants can be market-specific –
could also be found for the German market.
Chapter 5
The Impact of Model

Complexity on the Simulation
Results
Every valuation process is based on a broad set of assumptions. Examples are as-
sumptions regarding the input factors, the output factors, the relation between these
factors, and the necessary forecast models. In the case of stochastic simulation, ad-
ditional assumptions regarding the simulation procedure need to be taken. However,
it is rarely known which of the assumptions are crucial for the accuracy of the valu-
ation results. Therefore, the question posed in this chapter is regarding the impact
of certain assumptions on the valuation results.
In this chapter I quantify the impact of model complexity on the valuation results. I
identify two parameter categories which have a significant impact, namely the applied
forecast models and the general simulation setup. I valuate a power plant venture
which is financed via project finance to quantify the impact. However, the focus of
the case study is not on the precise specification of a power plant project but solely
on the estimation of the robustness of the valuation result in relation to certain
changes in the assumptions. At the beginning, I first give an introduction to the
case study, addressing the specifics of the project and the simplifying assumptions. I
define a base case as a parameter combination used at the beginning of the valuation
process. Later, I relax the assumptions and first vary the simulation parameters and
then the applied forecast models. The critical parameters and forecast models are
identified. I conclude the chapter with a summary of the results and an outline of
promising avenues for future research.
178
179
• What are the basics of a power plant venture financed via project finance and
which simplifying assumptions are justified?
• What is the impact of certain assumptions regarding the general simulation

setup on the valuation results, i.e. which parameters are crucial for the valua-
tion?
• What is the impact of certain assumptions regarding the applied forecast mod-
els on the valuation results, i.e. which parameters are crucial for the valuation?
In this chapter the PFVT is applied within a case study to determine the profitability
of a coal power plant financed via project finance. The research question underly-
ing the case study is the quantification of the impact of model complexity on the
valuation results.
The valuation is performed both from the equity and the debt provider’s point of
view. As valuation results I consider two main outputs, namely the probability dis-
tribution of the expected NPV and the cumulative default probability. Therefore,
the aimed variations in the input parameters are applied in the following for the cal-
culation of the NPV distribution and of the expected cumulative default probability
(ECDP).
I analyze the effect of model complexity on the NPV distribution and the (cumula-
tive) default probability of the project. In all simulations I first use the base case
defined in section 5.2.2. Then, in each step, one parameter is changed. I divide model
complexity into two components: (i) the complexity of the simulation procedure and
(ii) the complexity of the forecast models. The complexity of the simulation proce-
dure is defined along three dimensions: the number of iterations, the time-resolution
(which defines how often the cash flow is analyzed), and the valuation method for
equity. To explore their restrictive effects on the simulation results, I vary these three
components across several dimensions. For forecast complexity, I vary the volatil-
ity and the correlation forecast models. For volatility forecasts, I use (i) historical
volatility and forecasts obtained from (ii) a GARCH (1,1), (iii) an E-GARCH as
well as (iv) a GJR-model. Regarding correlations, I apply (i) no correlations, (ii)
historical correlations and (iii) correlations obtained from a DCC model.
Figure 5.1 summarizes the different variations of the model complexity which are
analyzed in the following.
180
Figure 5.1
Dimensions of Model Complexity
1.000
10.000
Iterations
100.000
500.000
Daily
Simulation Resolution Weekly
Complexity
Monthly
Equity Book Value

Value QMV
Model
Complexity
Historical
GARCH
Volatility
EGARCH
GJR
Forecast
Complexity
No Corr.
Correlation Historical
DCC
Source: Own work.
5.2 Case Study
5.2.1 Introduction
As a detailed consideration of all aspects of a real-life power plant project is beyond

the scope of this chapter, I apply several simplifications to ease the specification of
the project.1 Since I am not focused on the (absolute) results for this certain project,
but on the effects of model complexity, I expect these simplifications not to bias the
results.
I assume that the coal power plant’s construction takes place at one point-in-time,
that the whole capital is immediately invested, and that the power plant is operated
as a base load power plant. After the initial capital expenditure I identify four main
types of relevant costs for the power plant project:
1. fixed costs: e.g. salaries, insurances,
2. variable costs: e.g. chemicals,
3. fuel costs: costs for coal, and

1
These assumptions are made due to simplification matters and are not necessary for the PFVT.
The tool can handle very complex project structures.
181
4. costs for emission rights.
The power plant generates revenues which are dependent on the amount of electric-
ity produced and the electricity price. The computation of the amount of electricity
produced is not straightforward since several factors have to be taken into consider-
ation. These factors are (i) capacity, (ii) load factor, (iii) operating time, and (iv)
efficiency. To simplify matters, I assume that there is no technical improvement over
the project’s runtime; the efficiency of the power plant remains constant over the life-
time. Important factors affecting the project’s cash flow are the price of electricity,
the coal cost (coal input per MWh of electricity output), and the costs for emissions
rights (costs per MWh of electricity output). These factors are the main drivers of
the profitability of the power plant. As a detailed discussion of the underlying cash
flow equation is already provided in the previous chapter I refer the interested reader
to section 3.3.2.1.
I assume a capital requirement of 1 billion Euro as the initial investment for the build-
up phase of the power plant. The financial structure in the case study corresponds
to a typical project financed power plant. The debt-to-equity ratio at the beginning
of the project is two2 and the debt has a maturity of 20 years. Furthermore, I
assume that one third of the debt is provided in U.S. Dollar and two thirds are
provided in Euro. The runtime of the power plant is estimated at 40 years, with
total depreciation to be reached after 20 years.
The whole FCFE is immediately distributed among the sponsors. An event of default
generally occurs when the FCFE becomes negative, implying that the sponsors must
re-invest cash in the project. However, this assumption would be very restrictive and
not meaningful from an economic point of view. Thus, I assume that the sponsors
have the obligation to re-invest money into the project up to a threshold of one third
of their initial investments. Thereafter, every negative FCFE leads to an event of
default of the project. The capital structure of the project and its cost of equity is
permanently recalculated and adjusted during the simulated period of the project.3
Regarding the technical parameters of the power plant, an electricity generation
capacity of 1000 MW, a load factor of 69% and an availability of 84% are assumed.
Operating and maintenance costs are 5% of revenues plus 1 Euro for each generated
MWh. Fixed costs, which include labor costs, are assumed at 5 million Euro per
year. Regarding the relation between electricity output and fuel input I assume
that 0.978 emission rights are necessary for each generated MWh of electricity. In
2
Cf. section 2.1.1.2 for a discussion of usually observed debt-to-equity ratios in project finance
investments.
3
Cf. section 3.3.2.4 for a discussion of the NPV calculation.
182
addition, the case study assumes 36% thermal efficiency for the generation process.4
Furthermore, I assume that sponsors taking an equity stake in project finance typi-
cally have a limited investment horizon of three to seven years.5 By contrast, power
plant projects have a runtime of 20 years and more. As a consequence, a sponsor
must sell his stake in the project during its runtime. In order to do this, it is crucial
for the sponsor to be able to determine the value of his stake at any point-in-time.
The PFVT is able to simulate the project over its whole runtime. However, from
an economic point of view, a simulation over such a long time period does not make
much sense. To solve this issue, I model the project over a horizon of 5 years and as-
sume that the whole project is sold to another investor afterwards. The selling price
is calculated with the help of a valuation technique based on cash flow multiples, a
method often applied in real-life project finance investments.
The starting point of the analysis is May 2009. As mentioned above, I model the
first five years of the project’s runtime based on the stochastic modeling of the cash
flows. Afterwards, I assume that the project is sold for the price calculated using
a multiple based valuation approach. As common for power plant valuations, I
calculate its value as two times the average of its last three annualized free cash
flows to equity.6
5.2.2 Base Case
For the beginning of the valuation process I define a base case, a parameter and
forecast model setup used for the calculation of the first basic results. The following
forecasting methods are used for the different influencing factors within the base
case: I apply an ARMA(1,1) model for the electricity price, the emission rights, the
interest rates and the U.S. Dollar / Euro exchange rate. For the coal price I use an
ARMA(2,2) model.
The power plant is assumed to operate 24 hours and 7 days a week as it is natural
for base load power plants. This allows the forecasting of electricity prices without
consideration of their specifics, in particular their extreme intra-day and intra-week
variation.7 The reference price for electricity is the daily day-ahead market price from
the EEX as this is the market segments with the highest market liquidity. Moreover,
the use of futures prices is not possible as hedging is not yet implemented.8 All price
forecasts are based on the data analyzed in the previous chapter.
4
These assumptions are based on discussions with both practitioners and researchers. Further-
more data is taken from Wagner et al. (2004) and IEA (2010).
5
This assumption is based on discussions with project finance investors.
6
This assumption is based on discussions with project finance investors.
7
Cf. chapter 4 for a discussion of the characteristics of electricity prices.
8
It is intended to implement hedging and hence the use of futures prices in the near future.
183
Figure 5.2
Impact Number of Iterations on NPV
Source: Own work, based on Weber et al. (2010).
Further properties of the base case are:
1. all price forecasts are obtained based on ARMA-models9 ,
2. volatilities forecasts are based on a GARCH (1,1) model,
3. future correlations are assumed to be equal to historical correlations,
4. the model’s time resolution is weekly,
5. the calculation of the equity’s market value is based on the QMV-method, and
6. the number of iterations is 100,000.
5.3 Case Study: Results
5.3.1 Impact of Simulation Complexity
The number of iterations is the first parameter to be analyzed. It is expected that

an increase in this parameter leads to a smoother distribution function of the NPV.
9
I do not vary the price forecast models since their impact on the simulation outcome is obvious
184
Figure 5.3
Impact Number of Iterations on Default Probability
As expected, the level of the distribution function remains (largely) unchanged. Fig-
ure 5.2 presents the distribution function of simulations with different numbers of
iterations. I simulate the project with 1,000, 10,000, 100,000, and 500,000 iterations.
As expected, the number of iterations has no significant effect on the overall level
since there is no systematic shift in the cumulative distribution function. However,
for 1,000 and 10,000 iterations, the function is rather unsteady, as shown in the
enlargement of figure 5.2. As a consequence, the expected NPV may be over- or un-
derestimated, depending on whether the function is above or below its “true” value
at a certain point. The application of at least 100,000 of iterations seems to be favor-
able. The step from 100,000 to 500,000 iterations does not increase the smoothness
of the function significantly. I recommend using at least 100,000 iterations for a sim-
ulation to obtain smooth distribution functions. The same effects can be observed
for the ECDP, which is reported in figure 5.3.
The drawback of an increased number of iterations is that the computation time
rises significantly. As presented in figure 5.4, the computation time on the used
system, a commercially available high-end personal computer, rises from 2 minutes
for 1,000 iterations to 270 minutes for 500,000 iterations. Therefore, it is necessary
to weight the advances of an increased number of iterations, a smoother distribution
185
Figure 5.4
Impact Number of Iterations on Computation Time
2h
2h
2h
1h
1h
1h Complete
Monte Carlo
1h
1h
0h
0h
0h
1000 10000 100000 500000
function, against its drawback in terms of more computation time. As mentioned

before, an increase from 100,000 to 500,000 iterations does not lead to a strong
improvement in smoothness, but to a five times higher computation time. However,
below 100,000 iterations the distribution function is very rocky and not satisfying. As
a consequence, the above proposed 100,000 iterations seem to be a good compromise
between distributions smoothness and computation time.
Next, I analyze the impact of the applied equity valuation method on the simulation
results. Figure 5.5 presents the two methods included in the PFVT, equity valuation
based on QMV, and on book values.
I find that the cumulative distribution function differs for the two methods. The NPV
distribution is narrower with less extreme values for the equity valuation based on
book values compared to the QMV. The rationale behind this observation is that the
application of equity book values leads to biased estimates of the cost of equity since
they are not linked to the success of the project. In reality, the costs of equity depend
on the success and, hence, on the risk of a project. Since book value estimation does
not take this aspect into account, it overestimates the costs of equity for successful
projects and underestimates them for unsuccessful projects. The QMV links the costs
of equity to the profitability of a project to avoid this misjudgment. This estimation
186
Figure 5.5
Impact Equity Valuation Method on NPV
on the one hand leads to more extremely high NPV estimates (because the discount
rate for successful projects is smaller) and on the other hand to more extremely low
NPV values (because the discount rate for unsuccessful projects is higher). This can
be seen in the cumulative distribution function. The project’s ECDP is not affected
by the equity valuation method. Consequently, I do not report this figure.
The time resolution is another crucial parameter of the simulation. The time reso-
lution defines how often the project’s cash flow is analyzed. For example, a weekly
time resolution means that the cash flow of the project is computed and analyzed
each week during the simulation phase of the project. I analyze the impact of a
weekly, monthly, and a yearly time resolution. Figure 5.6 presents the impact of the
time resolution on the simulation results.
As it can be seen, a higher time resolution leads to a broader NPV distribution and
to more large positive and negative events. The higher frequency of small (negative)
NPV estimates is not surprising since a higher time resolution is accomplished by
more default events as presented in figure 5.7.
Since the cash flow is more often analyzed with, for example, weekly resolution
compared to yearly resolution, a stream of negative cash flows over several weeks
may lead to a project default. These events of default do not necessarily happen
187
Figure 5.6
Impact Time Resolution on NPV
Figure 5.7
Impact Time Resolution on Default Probability

188
Figure 5.8
Impact Volatility Forecast Model on NPV
with yearly resolution since a stream of negative cash flows may be followed by more
positive cash flows which compensate for the prior losses. However, on a weekly
resolution the project would have defaulted with no possibility of recovering. The
explanation of the higher frequency of large NPV estimates is more complicated.
One aspect that has to be taken into consideration is the effect of discounting, since
higher time resolution leads to smaller discounting steps. As a result, large positive
values in the future are discounted with a lower average discounting rate for high-time
frequencies, leading to higher NPV estimates today.
5.3.2 Impact of Forecast Complexity
The first issue to be addressed within the area of forecast complexity is how future
volatilities are predicted. The PFVT is able to either apply historical volatility values
as forecasts for future volatility or to compute those estimates based on different
GARCH type models. In addition to the GARCH(1,1) model, I also apply the more
advanced E-GARCH and GJR-model. An issue worth mentioning is that forecasts
based on GARCH models converge to forecasts based on historical values after a
certain period of time. The impact of volatility forecasts on the simulation results is
presented in figure 5.8.
189
Figure 5.9
Impact Volatility Forecast Model on Default Probability
As can be seen, the results depend heavily on the applied volatility forecast model.
The E-GARCH model leads to a much narrower NPV distribution compared to the
other models. The broadest NPV distribution is obtained with the GJR-GARCH
model. When analyzing the project’s ECDP, reported in figure 5.9, the same effects
can be found.
Especially the E-GARCH model is noticeable since it leads to a much lower default
probability. The gap between the default probability calculated by the E-GARCH
and the GJR-GARCH model is about 10% over the whole runtime of the project.
Evaluating which model is best is not possible since the “real” cumulative default
probability is unknown. However, my results suggest that the choice of the volatility
forecast model is extremely crucial for the simulation results. Hence, it is important
to figure out which volatility forecast model is best suited for each factor when valuing
a project in practice. The substantial impact of this choice can be seen from the prior
two figures. Recent academic works by Bowden & Payne (2008) and Chan & Gray
(2006) propose that E-GARCH is the best volatility model for electricity prices.
Hence, it could be argued that its application should be favored over alternative
models.
In a further step, I investigate the influence of the correlation forecast model. The
190
Figure 5.10
Impact Correlation Forecast Model on NPV
PFVT can compute future correlations in three different ways: (i) it assumes that the
correlation between all factors is zero in the future, (ii) historical correlations are used
as forecasts for future correlations, and (iii) estimates from a DCC model are applied
as predictors for correlations. The first method, which assumes zero correlations,
has to be interpreted with caution since fundamental economic relationships may
be neglected. For example, if a project is simulated whose cash flow depends on
gas and oil prices, the assumption of no correlation is clearly misleading since gas
and oil prices are highly correlated based on their linking mechanism. However, I
integrated this possibility to investigate whether the assumption of no correlation
has a significant impact on the results of the case study.10 Figure 5.10 presents the
cumulative NPV distribution function for these three different correlation forecasts.
As can be seen, the choice of the correlation forecast has a limited impact on the sim-
ulation results, especially when compared to the choice of the volatility model. The
simulated NPV distributions are relatively similar for all three correlation forecast
methods. The same is true for the ECDP, which is reported in figure 5.11.
To summarize, my results suggest that correlation forecasts based on a DCC model
10
As for GARCH based volatility forecasts, correlation forecasts based on the DCC model con-
verge to forecasts based on historical values after a certain time.
191
Figure 5.11
Impact Correlation Forecast Model on Default Probability
do not significantly improve the result compared to the application of historical

correlations. Taking into consideration that the DCC model is rather complicated
to implement in a valuation model, I suggest using historical correlations which are
easier to handle.
The impact of model complexity on the valuation results is quantified in this chapter.
I analyze whether model complexity matters for the valuation of project finance
investments in this chapter. The valuation is based on stochastic simulation and
performed by a newly developed project finance valuation tool which is introduced
in the previous chapter. I analyze the effect on the NPV distributions and the
estimated default probabilities. For this purpose, I apply the valuation tool on a
power plant case study. I distinguish between two dimensions of model complexity:
(i) the complexity of the simulation procedure and (ii) the complexity of the forecast
models.
The obtained results are as follows: Regarding the simulation parameters, I find
that considering the trade-off between result adequacy and computation time, a
192
number of 100,000 iterations seems to be the optimal choice. Furthermore, the

quasi-market valuation method should be used when valuing projects. Otherwise the
project’s cost of capital is either over- or understated, resulting in biased valuation
results. Analyzing the effect of the time resolution, I find that it is significant for the
calculation of the default probability. When using the typical text-book assumption
that the cash flow is received at the end of a period, the time resolution also has an
impact on the NPV distribution. This is due to the discounting of the individual
cash flows. When analyzing the effect of the forecast models, I find that the selected
volatility forecast model significantly affects the simulation results. The effect of the
correlation forecast model seems to be less pronounced.
To summarize, I am able to show in this chapter that model complexity is important
in the context of project finance valuation. However, there are model elements that
are extremely crucial while others seem to be less important.
Chapter 6
Summary and Conclusion
The goal of this dissertation is the development of a valuation model for project
finance in general and for power plant ventures financed via project finance in par-
ticular. Three questions that are essential for the development of the valuation model
are posed. The first question concerns the specific requirements for a computer-based
project finance valuation tool based on stochastic cash flow modeling and focused
on power plant ventures. The second question relates to the choice of the electricity
price time series for the calibration of the valuation model, i.e. to the specifics of
electricity prices and the German electricity wholesale market. The second and last
question relates to the impact of model complexity, i.e. the applied forecast models
and the precise simulation procedure, on the valuation results.In the following, I
discuss the obtained results in this dissertation for all three questions, putting the
results into context. In addition, based on the obtained results, implications for
practice and for future research are derived.
The first question addresses the development of a computer-based, self-contained
valuation tool. To achieve the stated goals a valuation tool is programmed in Matlab,
a powerful programming language for numerical applications. The tool is based on
stochastic cash flow modeling and uses various advanced forecast models. Forecasts
for level values, volatilities, and correlations are computed. The valuation tool is
designed to allow the user to choose which input factors are modeled as stochastic.
Furthermore, the tool is user-friendly to assure that it can also be used by non-trained
users. To increase the sophistication and the accuracy of the valuation process,
the modeling of correlation structures and non-analytical distributions for the input
factors is also implemented. The implementation of these extensions to standard
valuation tools significantly improves the theoretical foundation of the valuation
193
194
results. The main valuation results are probability distributions of future cash flows
and of the expected NPV. Moreover, the expected cumulative default probability is
estimated.
The second question arises from the lack of academic studies on the German electric-
ity market and in particular on the underlying theoretical price formation mechanism.
Thus, the choice of the right electricity price time series is not a trivial task as empir-
ical results for other markets imply the existence of risk premia in electricity futures
prices, i.e. these prices are not unbiased estimates of the expected spot prices. To
fill in this literature gap, I perform an analysis of the German electricity wholesale
market which is located at the EEX. Thereby, I apply the ex post approach. The
research question is whether empirical evidence for the the risk premia approach,
which is seen in the theoretical literature as the appropriate price formation mech-
anism, is present. I find significant but also mixed evidence for the existence of risk
premia. First, I conduct an analysis of the spot market. Due to the non-storability
of electricity, spot contracts are basically futures contracts. The data cover three
spot market segments, namely the intraday market, the block contract market, and
the day-ahead market; data range from August 2002 to June 2010. I find posi-
tive risk premia, both in the block contract market and in the day-ahead market.
However, the risk premia in the day-ahead market contracts are only significant on
non-working days. The risk premia vary in magnitude and in sign throughout the
day. I detect a term structure of risk premia during the sub-period in which all three
market segments were simultaneously existent and find that prices were higher in
market segments with a longer time-to-delivery. When testing for seasonality and a
time evolution of the risk premia, I find no significant results. Moreover, I have to
reject the hypothesis of a relationship between the risk premia and the spot price
variance and skewness. Second, I analyze the futures market, based on data rang-
ing from June 2002 to June 2010. Due to liquidity considerations I focus on month
futures, i.e. futures with a delivery period of one month. I find evidence for signif-
icant positive risk premia in short-term futures, i.e. futures with a time-to-delivery
up to around three months. When testing for a term structure of risk premia, the
results are mixed, since there are periods with increasing and decreasing risk premia.
Thus, the evidence suggests that the term structure is not stable. The existence of
seasonality in the risk premia cannot be tested with significant results; however, I
find evidence for higher risk premia in the winter months. When testing for factors
influencing the risk premia a relationship between the risk premia and the variance
and skewness of the spot prices has to be denied.
To answer the third question a case study is performed. The case study serves for
quantifying the impact of model complexity on the valuation results. I valuate a
195
power plant and analyze the impact on the NPV distribution and on the expected
default probability for certain parameter variations. Model complexity is analyzed
along two dimensions: simulation complexity and forecast complexity. I aim to iden-
tify model elements which are crucial for the valuation of project finance in practice
and thus vary several model aspects in order to analyze their impact on the valuation
results. For forecast complexity, I apply different volatility and correlation forecast
models, e.g. correlation forecasts based on both historical values and on a DCC
model. Regarding simulation complexity, the number of Monte Carlo iterations, the
equity valuation method, and the time resolution are varied. I find that the applied
volatility forecast models have a strong influence on the expected NPV distribution
and on the probability of default. In contrast, correlation forecast models play a
minor role. Time resolution and equity valuation are both crucial when specifying a
valuation model for project finance. Regarding the number of Monte Carlo iterations,
I demonstrate that 100,000 iterations are sufficient to obtain reliable results. Thus,
a realistic valuation task must focus on the appropriate forecast of the volatility and
the specification of the stochastic modeling.
Several implications for practice can be derived from the obtained results. First, the
newly developed valuation tool allows for the deriving of recommendations for the de-
velopment of other, more general valuation tools. As I identify various crucial aspects
of the valuation task, the focus within the development of a valuation tool should
be on these aspects. Moreover, in order to increase the computational performance
of the valuation task, the obtained results provide decision support for the appro-
priate balance between model complexity and simplifying assumptions. Second, the
obtained valuation results for the power plant venture have the potential to serve as
reference values for future valuation tasks. In particular the possibility to estimate
the ex ante default probability of the project, an advantage of the chosen modeling
approach based on Monte Carlo simulation, provides additional insight compared to
other modeling approaches. Third, the results on the price formation mechanism
in the German electricity wholesale market, i.e. the existence of risk premia, imply
that the futures prices are not unbiased estimates of the expected spot prices. This
has several implications for market participants regarding their hedging strategies,
the adjusting of investments plans to the expected prices, and the trading in certain
market segments. For instance, electricity consumers, i.e. markets participants in-
terested in holding a long position until the maturity of a future, should, based on
the results, take their position as early as possible in order to avoid paying a risk
premium. An electricity producer, on the other hand, should wait as long as possible
before selling its generated electricity. Regarding the choice of appropriate market
segments, market participants can use the obtained results to optimize their trading
strategies, such as trading in the intraday market, in particular on non-working days.
196
Based on the obtained results, future research on both the valuation of project fi-
nance and the price formation in the German electricity wholesale market also seems
promising.
The main avenue for future research on the valuation tool is examining whether
the obtained results hold true in a non-energy project finance context. Project
finance investments in other industries, e.g. telecommunication or infrastructure,
could be used to test the robustness of the results on model complexity. However, it is
difficult to obtain reliable data for those industries since little information is publicly
available. Furthermore, there are promising extensions that could be included in the
developed valuation tool, e.g. more sophisticated valuation methods as the real
options analysis and more advanced and realistic forecast models.
Regarding the price formation mechanism in the German electricity market, the
obtained results imply that future research in at least two directions seems to be
promising. First, further analysis on the time evolution of the risk premia is neces-
sary. Taking the point of view of an efficient market the magnitudes of the estimated
risk premia seem to be too high to be explained only with consideration of risk aver-
sion. Thus, parts of the academic literature conclude that the electricity markets are
not fully integrated with the broader financial markets. Other parts of the literature
conclude that the markets are still relatively young and an explanation based on
market fundamentals only is not sufficient. Further research is necessary, in partic-
ular on the question whether the risk premia decrease in magnitude with increasing
maturity of the market or rather an increasing level of market liquidity. The effect
of the entrance of industry outsiders, i.e. speculators, into the market is of interest.
Second, an identification of the fundamental drivers of the risk premia seems to be
promising. The role of fuels and of emission allowances are of particular interest
as recent academic results suggest a relationship between these measures and the
magnitude of the risk premia for other markets. However, it is always difficult to
generalize empirical results obtained for other electricity market, as every market has
its own characteristics which are determined by the underlying generation technology
mix.
Having provided answers to several fundamental questions and with promising av-
enues for future research, this dissertation represents a first step towards a deeper
understanding of the largest European electricity market, as well as of the valuation
of project finance investments. It aims to position the newly developed valuation tool
as a potential instrument for capital providers in their effort to accurately forecast
the cash flows and the corresponding value of these capital intensive investments,
which will shape the global infrastructure landscape for years to come.
Appendix A
Hourly Prices Intraday Market
The following six figures contain the time series of the 24 hour contracts in the in-
traday market on working days in the period January 2, 2008 to June 30, 2010.
Figure A.1
Time Series Intraday Market, Hour Contract 1 - 4
Hour 1 Hour 2
50 50
Price [Euro/MWH]
Price [Euro/MWH]
-50 -50
-150 -150
-250 -250
Hour 3 Hour 4
50 50
Price [Euro/MWH]
Price [Euro/MWH]
-50 -50
-150 -150
-250 -250
197
198
Figure A.2
Hour 5 Hour 6
100 100
Price [Euro/MWH]
Price [Euro/MWH]
0 0
-100 -100
-200 -200
Hour 7 Hour 8
100 100
Price [Euro/MWH]
Price [Euro/MWH]
0 0
-100 -100
-200 -200
Figure A.3
Hour 9 Hour 10
400 400
300 300
Price [Euro/MWH]
Price [Euro/MWH]
200 200
100 100
0 0
Hour 12
Hour 11
400
400
300
300
Price [Euro/MWH]
Price [Euro/MWH]
200
200
100
100
0
0
199
Figure A.4
Hour 13 Hour 14
400 400
300 300
Price [Euro/MWH]
Price [Euro/MWH]
200 200
100 100
0 0
Hour 15 Hour 16
400 400
300 300
Price [Euro/MWH]
Price [Euro/MWH]
200 200
100 100
0 0
Figure A.5
Hour 17 Hour 18
400 400
300 300
Price [Euro/MWH]
Price [Euro/MWH]
200 200
100 100
0 0
Hour 19 Hour 20
400 400
300 300
Price [Euro/MWH]
Price [Euro/MWH]
200 200
100 100
0 0
200
Figure A.6
Hour 21 Hour 22
200 200
150 150
Price [Euro/MWH]
Price [Euro/MWH]
100 100
50 50
0 0
Hour 23 Hour 24
200 200
150 150
Price [Euro/MWH]
Price [Euro/MWH]
100 100
50 50
0 0
Source: Own work.

Appendix B
Hourly Prices Day-Ahead

Market
The following six figures contain the time series of the 24 hour contracts in the day-
ahead market on working days in the period January 2, 2008 to June 30, 2010.
Figure B.1
Hour 1 Hour 2
100 100
0 0
Price [Euro/MWh]
Price [Euro/MWh]
-100 -100
-200 -200
Hour 3 Hour 4
100 100
0 0
Price [Euro/MWh]
Price [Euro/MWh]
-100 -100
-200 -200
201
202
Figure B.2
Hour 5 Hour 6
160 160
80 80
Price [Euro/MWh]
Price [Euro/MWh]
0 0
-80 -80
-160 -160
Hour 7 Hour 8
160 240
160
80 Price [Euro/MWh]
80
Price [Euro/MWh]
-80
-80
-160 -160
Figure B.3
Hour 9 Hour 10
300 300
200 200
Price [Euro/MWh]
Price [Euro/MWh]
100 100
0 0
Hour 11 Hour 12
300 300
200 200
Price [Euro/MWh]
Price [Euro/MWh]
100 100
0 0
203
Figure B.4
Hour 13 Hour 14
250 250
200 200
Price [Euro/MWh]
150
Price [Euro/MWh]
150
100 100
50 50
0 0
Hour 15 Hour 16
250 250
200 200
150 150
Price [Euro/MWh]
Price [Euro/MWh]
100 100
50 50
0 0
Figure B.5
Hour 17 Hour 18
600 600
400 400
Price [Euro/MWh]
Price [Euro/MWh]
200 200
0 0
Hour 19 Hour 20
600 600
400 400
Price [Euro/MWh]
Price [Euro/MWh]
200 200
0 0
204
Figure B.6
Hour 21 Hour 22
200 200
150 150
Price [Euro/MWh]
Price [Euro/MWh]
100 100
50 50
0 0
Hour 23 Hour 24
200 200
150 150
Price [Euro/MWh]
Price [Euro/MWh]
100 100
50 50
0 0
Source: Own work.

Bibliography
Andeweg, H., Dorsman, A. & von Montfort, K. (2009), ‘Electricity traffic over the
barriers of networks: The case of Germany and The Netherlands’, Frontiers in
Finance and Economics 6, 120–139.
Andren, N., Jankensgard, H. & Oxelheim, L. (2005), ‘Exposure-based cash-flow-at-

risk under macroeconomic uncertainty’, Working Paper No. 635, The Research
Institute of Industrial Economics .
Armtrong, M. & Galli, A. (2005), ‘Are day-ahead prices for electricity converging
in continental Europe? An exploratory data approach’, Working Paper, Ecole
Nationale Supérieure des Mines de Paris, February 2005 .
Backhaus, K., Lang, C. & Werthschulte, H. (2003), Evolution der Projekt-

finanzierungsstrukturen im Energiesektor am Beispiel der Stromerzeugung in
den USA, in K. Backhaus & H. Werthschulte, eds, ‘Projektfinanzierung.
Wirtschaftliche und rechtliche Aspekte einer Finanzierungsmethode für Grosspro-
jekte’, 2nd edn, Schäffer-Poeschel Verlag, Stuttgart.
Balcombe, K. G. & Smith, L. E. D. (1999), ‘Refining the use of Monte Carlo

techniques for risk analysis in project planning’, Journal of Development Stud-
ies 36, 113–135.
Bask, M. & Widerberg, A. (2009), ‘Market structure and the stability and volatility
of electricity prices’, Energy Economics 31, 278–288.
Bautista, G., Anjos, M. F. & Vannelli, A. (2007), ‘Modeling market power in electric-
ity markets: Is the devil only in the details?’, The Electricity Journal 20, 82–92.
Bauwens, L., Laurent, S. & Rombouts, J. V. (2006), ‘Multivariate GARCH models:

A survey’, Journal of Applied Econometrics 21, 79–109.
Beale, C., Chatain, M., Fox, N., Bell, S., Berner, J., Preminger, R. & Prins, J.
(2002), ‘Credit attributes of project finance’, Journal of Structured and Project
Finance 8, 5–9.
205
206
Benth, F. E., Cartea, A. & Kiesel, R. (2008), ‘Pricing forward contracts in power
markets by the certainty equivalence principle: Explaining the sign of the market
risk premium’, Journal of Banking and Finance 32, 2006–2021.
Benz, E. & Trück, S. (2006), ‘CO2 emission allowances trading in Europe - Specifying
a new class of assets’, Problems and Perspectives in Management 4, 30–40.
Bessembinder, H. (1992), ‘Systematic risk, hedging pressure, and risk premiums in

futures markets’, Review of Financial Studies 5, 637–667.
Bessembinder, H. & Chan, K. (1992), ‘Time-varying risk premia and forecastable

returns in futures markets’, Journal of Financial Economics 32, 169–193.
Bessembinder, H. & Lemmon, M. L. (2002), ‘Equilibrium pricing and optimal hedg-

ing in electricity forward markets’, Journal of Finance 57, 1347–1382.
Bierbrauer, M., Menn, C., Rachev, S. T. & Trück, S. (2007), ‘Spot and derivative
pricing in the EEX power market’, Journal of Banking and Finance 31, 3462–3485.
Black, F. (1976a), ‘The pricing of commodity contracts’, Journal of Financial Eco-

nomics 3, 167–179.
Black, F. (1976b), ‘Studies in stock price volatility changes’, Proceedings of the 1976
Meeting of the Business and Economic Statistics Section, American Statistical
Association pp. 177–181.
Bodie, Z. & Rosansky, V. I. (1980), ‘Risk and return in commodity futures’, Financial
Analysts Journal 36, 27–39.
Bollerslev, T. (1986), ‘Generalized autoregressive conditional heteroskedasticity’,

Journal of Econometrics 31, 307–327.
Bollerslev, T., Chou, R. Y. & Kroner, K. F. (1992), ‘ARCH modeling in finance’,

Journal of Econometrics 52, 5–59.
Bonetti, V., Caselli, S. & Gatti, S. (2010), ‘Offtaking agreements and how they
impact the cost of funding for project finance deals: A clinical case study of the
Quezon Power Ltd Co.’, Review of Financial Economics 19, 60–71.
Bonneville, E. & Rialhe, A. (2005), ‘Impact of liberalization of the electricity market

on energy efficiency, quality of supply and environmental performance’, Discussion
Paper, September 2005 .
Boogert, A. & Dupont, D. (2005), ‘On the effectiveness of the anti-gaming policy be-
tween the day-ahead and real-time electricity markets in The Netherlands’, Energy
Economics 27, 752–770.
207
Borenstein, S., Bushnell, J., Knittel, C. R. & Wolfram, C. (2008), ‘Inefficiencies and
market power in financial arbitrage: A study of California’s electricity markets’,
Journal of Industrial Economics 56, 347–378.
Borgonovo, E., Gatti, S. & Peccati, L. (2010), ‘What drives value creation in invest-
ment projects? An application of sensitivity analysis to project finance transac-
tions’, European Journal of Operations Research 205, 227–236.
Bosco, B., Parisio, L., Pelagatti, M. & Baldi, F. (2010), ‘Long-run relations in Eu-
ropean electricity prices’, Journal of Applied Econometrics 25, 805–832.
Böttcher, J. & Blattner, P. (2010), Projektfinanzierung, 2nd edn, Oldenbourg, Mu-

nich.
Botterud, A., Kristiansen, T. & Ilic, M. D. (2010), ‘The relationship between spot
and futures prices in the Nord Pool electricity market’, Energy Economics 32, 967–
978.
Bowden, N. & Payne, J. E. (2008), ‘Short term forecasting of electricity prices

for MISO hubs: Evidence from ARIMA-EGARCH models’, Energy Economics
30, 3186–3197.
Box, G. & Jenkins, G. (1970), Time Series Analysis - Forecasting and Control, 1st
edn, John Wiley, San Francisco.
Brealey, R. A., Myers, S. C. & Allen, F. (2007), Principles of Corporate Finance,

9th edn, McGraw-Hill Irvin, New York.
Brennan, M. J. (1958), ‘The supply of storage’, American Economic Review 48, 50–
72.
Brigham, E. F., Gapenski, L. C. & Daves, P. R. (1999), Intermediate Financial

Management, 6th edn, The Dryden Press, Orlando, Florida.
Brooks, C. (2002), Introductory Econometrics for Finance, 1st edn, Cambridge Uni-
versity Press, Cambridge.
Bryant, H. L., Bessler, D. A. & Haigh, M. S. (2006), ‘Causality in futures markets’,

Journal of Futures Markets 26, 1039–1057.
Bundesministerium für Wirtschaft und Technologie, B. (2009), Energie in Deutsch-

land - Trends und Hintergründe zur Energieversorgung, Aktualisierte Ausgabe,
April 2009, BMWi offentlichkeitsarbeit, Berlin.
208
Bundesministerium für Wirtschaft und Technologie, B. (2010), Energiedaten - Zahlen

und Fakten. Nationale und Internationale Entwicklung, Bundesministerium für
Wirtschaft und Technologie, Referat III C 3.
Bundesnetzagentur (2010), ‘Bundesnetzagentur ordnet Netzregelverbund für die

deutschen Stromnetze an’, Pressemitteilung, Bundesnetzagentur .
Bunn, D. & Karakatsani, N. (2008), ‘Forecasting electricity prices’, International

Journal of Forecasting 24, 764–785.
Capitan Herraiz, A. & Rodriguez Monroy, C. (2009), ‘Analysis of the efficiency of

the Iberian power futures market’, Energy Policy 37, 3566–3579.
Cartea, A. & Figueroa, M. G. (2005), ‘Pricing in electricity markets: A mean revert-

ing jump diffusion model with seasonality’, Applied Mathematical Finance 12, 313–
335.
Cartea, A. & Villaplana, P. (2008), ‘Spot price modeling and the valuation of elec-
tricity forward contracts: The role of demand and capacity’, Journal of Banking
& Finance 32, 2502–2519.
Carter, C. A. (1999), ‘Commodity futures markets: A survey’, Australian Journal of

Agricultural and Resource Economics 43, 209–247.
Carter, C. A., Rausser, G. C. & Schmitz, A. (1983), ‘Efficient asset portfolios and
the theory of normal backwardation’, Journal of Political Economy 91, 319–331.
Chan, K. F. & Gray, P. (2006), ‘Using extreme value theory to measure value-at-risk
for daily electricity spot prices’, International Journal of Forecasting 22, 283–300.
Chang, E. C. (1985), ‘Returns to speculators and the theory of normal backwarda-

tion’, Journal of Finance 40, 193–208.
Chatrath, A., Liang, Y. & Song, F. (1997), ‘Commitment of traders, basis behav-
ior, and the issue of risk premia in futures markets’, Journal of Futures Markets
17, 707–731.
Chiu, Y.-C. (2007), ‘Corporate diversification and risk management: Methodological

approach’, Journal of Financial Management and Analysis 20, 1–6.
Chow, Y.-F., McAleer, M. & Sequeira, J. M. (2000), ‘Pricing of forward and futures
contracts’, Journal of Economic Surveys 14, 215–253.
Considine, T. J. & Larson, D. F. (2001), ‘Risk premiums on inventory assets: The

case of crude oil and natural gas’, The Journal of Futures Markets 21, 109–126.
209
Convery, F. J. & Redmond, L. (2007), ‘Market and price developments in the Euro-
pean Union Emissions Trading Scheme’, Review of Environmental Economics and
Policy 1, 88–111.
Cootner, P. H. (1960), ‘Returns to speculators: Telser versus Keynes’, Journal of

Political Economy 68, 396–404.
Copeland, T. E. & Antikarov, V. (2001), Real Options - A practicioner’s guide, 1st

edn, Texere, London.
Copeland, T. E., Weston, J. F. & Shastri, K. (2005), Financial Theory and Corporate
Policy, 4th edn, Pearson Addison Wesley, Boston, New York.
Copeland, T., Koller, T. & Murrin, J. (2000), Valuation: Measuring and Managing
the Value of Companies, 3rd edn, John Wiley, New York.
Corielli, F., Gatti, S. & Steffanoni, A. (2008), ‘Risk shifting through nonfinancial
contracts. Effects on loan spreads and capital structure of project finance deals’,
Working Paper, Università Commerciale Luigi Bocconi, May 2008 .
Culp, C. & Forrester, J. (2010), Structured Financing Techniques in Oil & Gas
Project Finance: Future Flow Securitizations, Prepaids, Volumetric Production
Payments, and Project Finance Collateralized Debt Obligationsv, in A. Kramer &
P. Fusaro, eds, ‘Energy & Environmental Project Finance Law & Taxation: New
Investment Techniques’, 1st edn, Oxford University Press,.
Dailami, M. & Hauswald, R. (2001), ‘Contract risks and credit spread determinants
in the international project bond market’, World Bank Policy Research Working
Paper No. 2712, November 2001 .
Dailami, M., Lipkovich, I. & Van Dyck, J. (1999), ‘INFRISK. A computer simulation
approach to risk management in infrastructure project finance transactions’, World
Bank, Working Paper No. 2083 .
Damodaran, A. (1994), Damodaran on Valuation: Security Analysis for Investment

and Corporate Finance, 1st edn, John Wiley, New York.
Daskalakis, G. & Markellos, R. N. (2008), ‘Are the European carbon markets effi-
cient?’, Review of Futures Markets 17, 103–128.
Daskalakis, G. & Markellos, R. N. (2009), ‘Are electricity risk premia affected by

emission allowance prices? Evidence from the EEX, Nord Pool and Powernext’,
Energy Policy 37, 2594–2604.
210
Daskalakis, G., Psychoyios, D. & Markellos, R. N. (2009), ‘Modeling CO2 emission

allowance prices and derivatives: Evidence from the European trading scheme’,
Journal of Banking and Finance 33, 1230–1241.
de Roon, F. A., Nijman, T. E. & Veld, C. (2000), ‘Hedging pressure effects in futures
markets’, Journal of Finance 55, 1437–1456.
Decker, C. (2008), Internationale Projektfinanzierung: Konzeption und Prüfung, 1st

edn, Books on Demand, Norderstedt.
Deng, S. S. & Oren, S. S. (2006), ‘Electricity derivatives and risk management’,

Energy 31, 940–953.
Dennis, S. A. & Mullineaux, D. J. (2000), ‘Syndicated loans’, Journal of Financial

Intermediation 9, 404–426.
Diebold, F. (2004), ‘The Nobel Memorial Prize for Robert F. Engle’, Scandinavian
Journal of Economics 106, 165–185.
Dieckmann, B. (2008), Engpassmanagement im Europäischen Strommarkt, PhD the-

sis, Wesfälische Wilhelms-Universität Münster.
Diko, P., Lawford, S. & Limpens, V. (2006), ‘Risk premia in electricity forward
prices’, Studies in Nonlinear Dynamics and Econometrics 10, Article No. 7.
Dixit, A. K. & Pindyck, R. S. (1994), Investment under Uncertainty, 1st edn, Prince-
ton University Press, Princeton, New Jersey.
Douglas, S. & Popova, J. (2008), ‘Storage and the electricity forward premium’,
Energy Economics 30, 1712–1727.
Dusak, K. (1973), ‘Futures trading and investor returns: An investigation of com-

modity market risk premiums’, Journal of Political Economy 81, 1387–1406.
EEX (2008a), EEX Product Information Power. Leipzig, July 2008.
EEX (2008b), Introduction to Exchange Trading at EEX on XETRA and Eurex.

Leipzig, January 2008.
EEX (2009a), Geschäftsbericht 2008. Leipzig, 2009.
EEX (2009b), Index Description. Leipzig, April 2009.
EEX (2010a), EEX Exchange Rules. Leipzig, January 2010.
EEX (2010b), Unternehmen und Produkte. Leipzig, January 2010.

211
Ehrhardt, M. (1994), The Search for Value: Measuring the Company’s Cost of Cap-
ital, 1st edn, Harvard Business School Press, Boston, Massachusetts.
Ellerman, A. D. & Buchner, B. K. (2007), ‘The European Union Emissions Trading

Scheme: Origins, allocation, and early results’, Review of Environmental Eco-
nomics and Policy 1, 66–87.
Ellersdorfer, I., Hundt, M., Sun, N. & Voss, A. (2008), ‘Empirische Analysen der
Preisbildung am deutschen Elektrizitätsmarkt - Der Teufel steckt im Detail!’, En-
ergiewirtschaftliche Tagesfragen 58, 32–40.
Engle, R. (2002), ‘Dynamic conditional correlation: A simple class of multivariate

generalized autoregressive conditional heteroskedasticity models’, Journal of Busi-
ness & Economic Statistics 20, 339–350.
Engle, R. F. (1982), ‘Autoregressive conditional heteroskedasticity with estimates of

the variance of UK inflation’, Econometrica 50, 987–1008.
Engle, R. & Sheppard, K. (2001), ‘Theoretical and empirical properties of Dynamic

Conditional Correlation multivariate GARCH’, NBER Working Papers 8554, De-
cember 2001 .
Erner, C., Olinger, D. & Werthschulte, H. (2003), Einsatz des Realoptionsansatzes

zur Bewertung von Projektfinanzierungsvorhaben, in K. Backhaus & H. Werth-
schulte, eds, ‘Projektfinanzierung. Wirtschaftliche und rechtliche Aspekte einer
Finanzierungsmethode für Grossprojekte’, 2nd edn, Schäffer-Poeschel Verlag,
Stuttgart.
Escribano, A., Peña, J. I. & Villaplana, P. (2002), ‘Modeling electricity prices: In-
ternational evidence’, Working Paper, Universidad Carlos III de Madrid Working
Papers, June 2002 .
Esty, B. C. (1999), ‘Improved techniques for valuing large-scale projects’, Journal of

Project Finance 5, 9–26.
Esty, B. C. (2003), ‘The economic motivations for using project finance’, Working
Paper, Harvard Business School, February 2003 .
Esty, B. C. (2004), ‘Why study large projects? An introduction to research on project

finance’, European Financial Management 10, 213–224.
Esty, B. C. & Megginson, W. L. (2001), ‘Legal risk as a determinant of syndicate

structure in the project finance loan market’, Working Paper, Harvard Business
School, November 2001 .
212
Esty, B. C. & Sesia, A. J. (2007), ‘An overview of project finance and infrastructure
finance - 2006 update’, Harvard Business School Publishing Case No. 207-107 .
Esty, B. C. & Sesia, A. J. (2010), ‘An overview of project finance and infrastructure
finance - 2009 update’, Harvard Business School Publishing Case No. 210-061 .
Fahrholz, B. (1998), Neue Formen der Unternehmensfinanzierung, 1st edn, C.H.

Beck, Munich.
Fama, E. F. (1965), ‘Random walks in stock market prices’, Financial Analysts

Journal 21, 55–59.
Fama, E. F. & French, K. R. (1987), ‘Commodity futures prices: Some evidence on

forecast power, premiums, and the theory of storage’, Journal of Business 60, 55–
73.
Fama, E. F. & French, K. R. (2004), ‘The capital asset pricing model: Theory and
evidence’, Journal of Economics Perspectives 18, 25–46.
Fernandez, P. (2009), ‘Valuing companies by cash flow discounting: 10 methods and

9 theories’, Working Paper, IESE Business School, November 2009 .
Finnerty, J. D. (2007), Project financing: Asset-based financial engineering, 2nd edn,

John Wiley, New York.
Flinkerbusch, K. & Heuterkes, M. (2010), ‘Cost reduction potentials in the German

market for balancing power’, Energy Policy 38, 4712–4718.
Flyvbjerg, B., Holm, M. S. & Buhl, S. (2002), ‘Underestimating costs in public works
projects’, Journal of the American Planning Association 68, 279–295.
Fridolfsson, S.-O. & Tangera, T. P. (2009), ‘Market power in the Nordic electricity
wholesale market: A survey of the empirical evidence’, Energy Policy 37, 3681–
3692.
Frondel, M., Ritter, N., Schmidt, C. M. & Vance, C. (2010), ‘Economic impacts from
the promotion of renewable energy technologies: The German experience’, Energy
Policy 38, 4048–4056.
Frondel, M., Schmidt, C. M. & Vance, C. (2008), ‘Emissions trading: Impact on

electricity prices and energy-intensive industries’, Ruhr Economic Papers No. 81,
Rheinisch-Westfälisches Institut für Wirtschaftsforschung .
Furio, D. & Lucia, J. J. (2009), ‘Congestion management rules and trading strategies
in the Spanish electricity market’, Energy Economics 31, 48–60.
213
Furio, D. & Meneu, V. (2010), ‘Expectations and forward risk premium in the Span-
ish deregulated power market’, Energy Policy 38, 784–793.
Gatti, S., Kleimeier, S., Megginson, W. L. & Steffanoni, A. (2008), ‘Arranger certi-
fication in project finance’, Working Paper, July 2008 .
Gatti, S., Rigamonti, A., Saita, F. & Senati, M. (2007), ‘Measuring value-at-risk in
project finance transactions’, European Financial Management 13, 135–158.
Geman, H. (2005), Commodities and Commodity Derivatives: Modelling and Pricing

for Agriculturals, Metals and Energy, 1st edn, John Wiley, New York.
Geman, H. & Vasicek, O. (2001), ‘Plugging into electricity’, Risk August issue
2001, 93–97.
Ghersi, H. & Sabal, J. (2006), ‘An introduction to project finance in emerging mar-
kets’, Working Paper, ESADE Barcelona .
Glosten, L. R., Jagannathan, R. & Runkle, D. (1993), ‘On the relation between the
expected value and the volatility of the nominal excess return on stocks’, Journal
of Finance 48, 1779–1801.
Glynn, P. W. & Witt, W. (1992), ‘Asymptotic efficiency of simulation estimators’,

Operations Research 40, 505–520.
Graham, J. & Harvey, C. (2001), ‘The theory and practice of corporate finance:
Evidence from the field’, Journal of Financial Economics 60, 187–243.
Grimm, V., Ockenfels, A. & Zöttl, G. (2008), ‘Ein Vergleich ausgewählter europäis-
cher Strombörsen’, Zeitschrift für Energiewirtschaft 3, 162–170.
Grinblatt, M. & Titman, S. (2001), Markets and Corporate Strategy, 2nd edn, Irwin
McGraw-Hill, Boston.
Habib, M. A. & Johnsen, D. B. (1999), ‘The financing and redeployment of specific

assets’, The Journal of Finance 54, 693–720.
Hacura, A., Jadamus-Hacura, M. & Kocot, A. (2001), ‘Risk analysis in investment

appraisal based on the Monte Carlo simulation technique’, European Physical Jour-
nal B 20, 551–553.
Hadsell, L. (2008), ‘Day-ahead premiums on the New England ISO’, Electricity Jour-
nal 21, 51–57.
214
Hadsell, L. & Shawky, H. A. (2007), ‘One-day forward premiums and the impact of
virtual bidding on the New York wholesale electricity market using hourly data’,
Journal of Futures Markets 27, 1107–1125.
Hainz, C. & Kleimeier, S. (2004), ‘Political risk in syndicated lending: Theory and
empirical evidence regarding the use of project finance’, Working Paper, University
of Munich, January 2004 .
Hainz, C. & Kleimeier, S. (2006), ‘Project finance as a risk-management tool in

international syndicated lending’, Discussion Paper, GESY .
Hammami, M., Ruhashyankiko, J.-F. & Yehoue, E. B. (2006), ‘Determinants of

public-private partnerships in infrastructure’, Working Paper, International Mon-
etary Fund, April 2006 .
Hammersley, J. M. & Handscomp, D. C. (1964), Monte Carlo Methods, 1st edn,

Methuen, London.
Hartzmark, M. L. (1987), ‘Returns to individual traders of futures: Aggregate re-

sults’, Journal of Political Economy 95, 1292–1306.
Hartzmark, M. L. (1991), ‘Luck versus forecast ability: Determinants of trader per-

formance in futures markets’, Journal of Business 64, 49–74.
Hertz, D. B. (1964a), ‘Investment policies that pay off’, Harvard Business Review
46, 96–108.
Hertz, D. B. (1964b), ‘Risk analysis in capital investment’, Harvard Business Review

42, 95–106.
Hess, S. W. & Quigley, H. A. (1963), ‘Analysis of risk in investments using Monte

Carlo techniques’, Chemical Engineering Symposium Series 42: Statistical and
Numerical Methods in Chemical Engineering pp. 55–71.
Hicks, J. R. (1939), Value and Capital, 1st edn, Oxford University Press, New York.
Hirshleifer, D. (1988), ‘Residual risk, trading costs, and commodity futures risk
premia’, Review of Financial Studies 1, 173–193.
Hirshleifer, D. (1989), ‘Determinants of hedging and risk premia in commodity fu-

tures markets’, Journal of Financial and Quantitative Analysis 24, 313–331.
Hughes, W. T. (1995), ‘Risk analysis and asset valution: A Monte Carlo simulation
using stochastic rents’, Journal of Real Estate Finance and Economics 11, 177–
187.
215
Hui, Y. V., Leung, L. C. & Huang, J.-S. (1993), ‘Analysis of correlated risky cash
flow’, International Journal of Production Economics 32, 269–276.
Hull, J. C. (2008), Options, Futures, and other Derivatives, 7th edn, Pearson Edu-
cation, Upper Saddle River.
Hwee, N. G. & Tiong, R. L. K. (2002), ‘Model on cash flow forecasting and risk anal-
ysis for contracting firms’, International Journal of Project Management 20, 351–
363.
IEA, i. (2009), World Energy Outlook 2009 Fact Sheet. Why is our current energy
pathway unsustainable. International Energy Agency.
IEA, i. (2010), Projected costs of generating electricity. 2010 edition. International

Energy Agency.
Jagannathan, R. (1985), ‘An investigation of commodity futures prices using the

consumption-based intertemporal capital asset pricing model’, Journal of Finance
40, 175–191.
Janssen, M. & Wobben, M. (2009), ‘Electricity pricing and market power - Evidence
from Germany’, European Transactions on Electrical Power 19, 591–611.
Jensen, M. C. (1986), ‘Agency cost of free cash flow, corporate finance, and
takeovers’, American Economic Review 76, 323–329.
Jensen, M. C. & Meckling, W. H. (1976), ‘Theory of the firm: Managerial behavior,

agency costs and ownership structure’, Journal of Financial Economics 3, 305–
360.
Kaka, A. P. (1996), ‘Towards more flexible and accurate cash flow forecasting’, Con-
struction Management and Economics 14, 35–44.
Kaldor, N. (1939), ‘Speculation and economic stability’, Review of Economic Studies

7, 1–27.
Kaplan, S. N. & Ruback, R. S. (1995), ‘The valuation of cash flow forecasts: An

empirical analysis’, Journal of Finance 50, 1059–1093.
Karakatsani, N. V. & Bunn, D. W. (2005), ‘Diurnal reversals of electricity forward

premia’, Working Paper, London Business School, February 2005 .
Keynes, J. M. (1923), ‘Some aspects of commodity markets’, Manchester Guardian

Commercial, European Reconstruction Series pp. 784–786.
Keynes, J. M. (1930), A Treatise on Money, Macmillan, London.

216
Kilic, M. & Huisman, R. (2010), ‘Is power production flexibility a substitute for
storability? Evidence from electricity futures prices’, Working Paper, Tinbergen
Institute, July 2010 .
Kleimeier, S. & Megginson, W. L. (2001), ‘An empirical analysis of limited recourse

project finance’, Working Paper, University of Oklahoma, July 2001 .
Kleimeier, S. & Versteeg, R. (2009), ‘Project finance as a driver of economic growth in

low-income countries’, Working Paper, Maastricht Research School of Economics
of Technology and Organization .
Klompjan, R. & Wouters, M. J. F. (2002), ‘Default risk in project finance’, Journal

of Structured and Project Finance 8, 10–22.
Knittel, C. R. & Roberts, M. R. (2005), ‘An empirical examination of restructured

electricity prices’, Energy Economics 27, 791–817.
Kolb, R. W. (1996), ‘The systematic risk of futures contracts’, Journal of Futures

Markets pp. 631–654.
Kolos, S. P. & Ronn, E. I. (2008), ‘Estimating the commodity market price of risk
for energy prices’, Energy Economics 30, 621–641.
Kramer, N. (2002), Modellierung von Preisbildunsmechanismem im liberalisierten

Strommarkt, PhD thesis, Technische Universität Bergakademie Freiberg.
Krisp, A. (2007), Der deutsche Strommarkt in Europa - Zwischen Wettbewerb und

Klimaschutz, PhD thesis, Justus-Liebig-Universität Giessen.
Kruger, J., Oates, W. E. & Pizer, W. A. (2007), ‘Decentralization in the EU emissions

trading scheme and lessons for global policy’, Review of Environmental Economics
and Policy 1, 112–133.
Kwak, Y. H. & Ingall, L. (2007), ‘Exploring Monte Carlo simulation applications for
project management’, Risk Management 9, 44–57.
Lessard, D. & Miller, R. (2001), ‘Understanding and managing risks in large en-
gineering projects’, Working Paper, MIT Sloan School of Management, October
2001 .
Leuthold, R. M., Philip, G. & Lu, R. (1994), ‘The returns and forecasting ability
of large traders in the frozen pork bellies futures market’, Journal of Business
67, 459–473.
Li, L. H. (2000), ‘Simple computer applications improve the versatility of discounted

cash flow analysis’, The Appraisal Journal 68, 86–92.
217
Lintner, J. (1965), ‘The valuation of risk assets and the selection of risky investments
in stock portfolios and capital budgets’, Review of Economics and Statistics 47, 13–
37.
Longstaff, F. A. & Wang, A. W. (2004), ‘Electricity forward prices: A high-frequency

empirical analysis’, Journal of Finance 59, 1877–1900.
Lucia, J. J. & Torro, H. (2008), ‘Short-term electricity futures prices: Evidence

on the time-varying risk premium’, Working Paper No. 2008-08, Universitat de
València, February 2008 .
Mandelbrot, B. (1963), ‘The variation of certain speculative prices’, Journal of Busi-

ness 36, 394–419.
Marckhoff, J. (2009), ‘The valuation of long-term exchange options in the German

electricity market’, Working Paper, University of Bamberg, May 2009 .
Marckhoff, J. & Wimschulte, J. (2009), ‘Locational price spreads and the pricing
of contracts for difference: Evidence from the Nordic market’, Energy Economics
31, 257–268.
Markowitz, H. (1952), ‘Portfolio selection’, The Journal of Finance 7, 77–91.
Marsaglia, G. & Tsang, W. W. (1984), ‘A fast, easily implemented method for sam-
pling from decreasing or symmetric unimodal density functions’, SIAM Journal
on Scientific and Statistical Computing 5, 349–359.
Marsaglia, G. & Zaman, A. (1991), ‘A new class of random number generators’, The
Annals of Applied Probability 1, 462–480.
Marti, S. & Keith, L. (2000), ‘Cash flow volatility as opportunity: Adding sophis-
ticated insurance capital to the project finance mix’, Journal of Project Finance
6, 9–14.
Meeus, L., Belmans, R. & Glachant, J.-M. (2006), ‘Regional electricity market in-
tegration France-Belgium-Netherlands’, Revue et Société Royale Belge des Elec-
triciencs 122, 17–23.
Menzel, H.-B. (2007), Rolle und Bedeutung von Strombörsen im Europäischen Bin-
nenmarkt. Workshop "Transparenter Strommarkt - Utopie oder Wirklichkeit?!",
Präsentation Bundesnetzagentur, 26.11.07.
Metropolis, N. (1987), ‘The beginning of the Monte Carlo method’, Los Alamos
Science pp. 125–130.
218
Metropolis, N. & Ulam, S. (1949), ‘The Monte Carlo method’, Journal of the Amer-
ican Statistical Association 44, 335–341.
Modigliani, F. & Miller, M. (1958), ‘The cost of capital, corporate finance and in-
vestment’, American Economic Review 48, 655–669.
Mossin, J. (1966), ‘Equilibrium in a capital asset market’, Econometrica 34, 768–783.
Möst, D. & Genoese, M. (2009), ‘Market power in the German wholesale electricity
market’, The Journal of Energy Markets 2, 47–74.
Müller-Merbach, J. (2009), Bewertung von Termingeschäften auf Elektrizität, PhD

thesis, Universität Mannheim.
Müsgens, F. (2006), ‘Quantifying market power in the German wholesale electricity

market using a dynamic multi-regional dispatch model’, The Journal of Industrial
Myers, S. C. & Maljuf, N. S. (1984), ‘Corporate financing and investment decisions

when firms have information that investors do not have’, Journal of Financial
Nakamura, M., Nakashimab, T. & Niimurac, T. (2006), ‘Electricity markets volatil-

ity: Estimates, regularities and risk management applications’, Energy Policy
34, 1736–1749.
Nawrocki, D. (2001), ‘The problems with Monte Carlo simulation’, Journal of Fi-
nancial Planning 14, 92–106.
Nelson, D. B. (1991), ‘Conditional heteroskedasticity in asset returns: A new ap-

proach’, Econometrica 59, 347–370.
Neusser, K. (2009), Zeitreihenanalyse in den Wirtschaftswissenschaften, 2nd edn,

Vieweg+Teubner, Wiesbaden.
Nevitt, P. & Fabozzi, F. (2000), Project Finance, 7th edn, Euromoney Books, Lon-
don.
Nicolosi, M. (2010), ‘Wind power integration, negative prices and power system
flexibility - an empirical analysis of extreme events in Germany’, Working Paper,
Institute of Energy Economics at the University of Cologne, March 2010 .
Nicolosi, M., Fürsch, M. & Lindenberger, D. (2010), ‘Analyse der Ursachen für
negative Strompreise am 3./4. Oktober 2009 und möglicher Abhilfemassnahmen’,
Studie im Auftrag des Bundesministeriums für Wirtschaft und Technologie, En-
ergiewirtschaftliches Institut an der Universität zu Köln .
219
Ockenfels, A., Grimm, V. & Zoettl, G. (2008), Strommarktdesign. Preisbildunsmech-

anismus im Auktionsverfahren für Stundenkontrakte an der EEX. Gutachten im
Auftrag der European Energy Exchange AG zur Vorlage an die Sächsische Börse-
naufsicht, März 2008.
Pena, D., Tiao, G. C. & Tsay, R. S., eds (2001), A course in time series analysis,
1st edn, John Wiley, New York.
Penman, S. H. (2007), Financial Statement Analysis and Security Valuation, 3rd

edn, McGraw-Hill Irwin, New York.
Pichler, P. & Wilhelm, W. (2001), ‘A theory of the syndicate: Form follows function’,
The Journal of Finance 56, 2237–2264.
Pietz, M. (2009a), ‘Risk premia in athe German electricity futures market’, Working
Paper, Technische Universität München, May 2009 .
Pietz, M. (2009b), ‘Risk premia in electricity wholesale spot markets - Empirical evi-
dence from Germany’, Working Paper, Technische Universität München, Septem-
ber 2009 .
Pirrong, C. & Jermakyan, M. (2008), ‘The price of power: The valuation of power
and weather derivatives’, Journal of Banking and Finance 32, 2520–2529.
Polo, M. & Scarpa, C. (2002), ‘The liberalization of energy markets in Europe and
Italy’, Working Paper, Bocconi University, December 2002 .
Przybilla, A. (2008), Projektfinanzierungen im Rahmen des Risikomanagements, 1st

edn, Salzwasser Verlag.
Rajan, R. & Winton, A. (1995), ‘Covenants and collateral as incentives to monitor’,

The Journal of Finance 50, 1113–1146.
Rammerstorfer, M. & Wagner, C. (2009), ‘Reforming minute reserve policy in Ger-

many: A step towards efficient markets?’, Energy Policy 37, 3513–3519.
Redl, C., Haas, R., Huber, C. & Böhm, B. (2009), ‘Price formation in electricity
forward markets and the relevance of systematic forecast errors’, Energy Economics
31, 356–364.
Ronn, E. I. & Wimschulte, J. (2009), ‘Intra-day risk premia in European electricity

forward markets’, Journal of Energy Markets 2, 71–98.
Ross, S. A. (1976), ‘The arbitrage theory of capital asset pricing.’, Journal of Eco-
nomic Theory 13, 341–360.
220
Ross, S. A., Westerfield, R. W. & Jaffe, J. (2005), Corporate Finance, 7th edn,
McGraw-Hill, New York.
Röver, J.-H. (2003), Projektfinanzierung, in K. Backhaus & H. Werthschulte,

eds, ‘Projektfinanzierung. Wirtschaftliche und rechtliche Aspekte einer Fi-
nanzierungsmethode für Grossprojekte’, 2nd edn, Schäffer-Poeschel Verlag,
Stuttgart.
Samuelson, P. A. (1965), ‘Proof that properly anticipated prices fluctuate randomly.’,

Industrial Management Review 6, 41–49.
Saravia, C. (2003), ‘Speculative trading and market performance: The effect of ar-
bitrageurs on efficiency and market power in the New York electricity market’,
Working Paper, University of California, November 2003 .
Schill, W.-P. & Kemfert, C. (2009), ‘The effect of market power on electricity storage
utilization - The case of pumped hydro storage in Germany’, Working Paper, DIW
Berlin, November 2009 .
Schnorrenberg, B. (2006), Zur Preisbildung von Forwardkontrakten im Strommarkt,

PhD thesis, Universität zu Köln.
Schwartz, E. S. (1997), ‘The stochastic behavior of commodity prices: Implications

for valuation and hedging’, Journal of Finance 52, 923–973.
Schwarz, H.-G. & Lang, C. (2006), ‘The rise in German wholesale electricity
prices: Fundamental factors, exercise of market power, or both?’, Working Pa-
per, Friedrich-Alexander-Universität Erlangen-Nürnberg, August 2006 .
Schwarz, H.-G., Lang, C. & Meier, S. (2007), ‘Market power in the German wholesale
electricity market: What are the political options?’, Working Paper, Universität
Erlangen-Nürnberg, April 2007 .
Seifert, J. & Uhrig-Homburg, M. (2007), ‘Modelling jumps in electricity prices: The-

ory and empirical evidence’, Review of Derivatives Research 10, 59–85.
Sharpe, W. F. (1964), ‘Capital asset prices: A theory of market equilibrium under

conditions of risk’, Journal of Finance 19, 425–442.
Shawky, H. A., Marathe, A. & Barrett, C. L. (2003), ‘A first look at the empirical
relation between spot and futures electricity prices in the United States’, Journal
of Futures Markets 23, 931–955.
Sijm, J., Neuhoff, K. & Chen, Y. (2006), ‘CO2 cost pass through and windfall profits
in the power sector’, Working Paper, Energy Research Centre of the Netherlands,
University of Cambridge, The Johns Hopkins University, May 2006 .
221
Simons, K. (1993), ‘Why do banks syndicate loans?’, New England Economic Review
pp. 45–52.
Smith, D. J. (1994), ‘Incorporating risk into capital budgeting decisions using simu-
lation’, Management Decision 32, 20–26.
Soennecken, A., Niessen, S., Pilgram, T., Croonen, M. & Rütze, M. (2002), ‘OTC-
Clearing der EEX - neben dem Futuresmarkt eine Ausschlussvariante des Kred-
itrisikos beim Energiehandel’, Energiewirtschaftliche Tagesfragen 52, 760–763.
Soennecken, A. & Pilgram, T. (2002), ‘European Energy Exchange - Die neue En-
ergiebörse in Leipzig’, Energiewirtschaftliche Tagesfragen 52, 396–399.
Sorge, M. (2004), ‘The nature of credit risk in poject finance’, Bank for International
Settlements Quarterly Review pp. 91–101.
Sorge, M. & Gadanecz, B. (2008), ‘The term structure of credit spreads in project
finance’, International Journal of Finance and Economics 13, 68–81.
Spillers, C. A. (1999), ‘Implications of economic turmoil to emerging market project

finance transactions - A rating agency perspective’, Journal of Project Finance
4, 9–13.
Spinney, P. J. & Watkins, G. C. (1996), ‘Monte Carlo simulation techniques and

electric utility resource decisions’, Energy Policy 24, 155–163.
Stoll, H. R. (1979), ‘Commodity futures and spot price determination and hedging
in capital market equilibrium’, Journal of Financial and Quantitative Analysis
14, 873–894.
Subramanian, K., Tung, F. & Wang, X. (2007), ‘Project finance versus corprate
finance’, Working Paper, April 2007 .
Sufi, A. (2007), ‘Information asymmetry and financing arrangements: Evidence from

syndicated loans’, The Journal of Finance 62, 629–668.
Telser, L. G. (1958), ‘Futures trading and the storage of cotton and wheat’, Journal
of Political Economy 66, 233–255.
Thomson Reuters, T. (2009), Global project finance review, Technical report, Thom-
son Reuters, Fourth Quarter 2009.
Tsay, R. S. (2002), Analysis of financial time series, 2nd edn, John Wiley, New York.
222
Tytko, D. (2003), Grundlagen der Projektfinanzierung, in K. Backhaus & H. Werth-

schulte, eds, ‘Projektfinanzierung. Wirtschaftliche und rechtliche Aspekte einer
Finanzierungsmethode für Grossprojekte’, 2nd edn, Schäffer-Poeschel Verlag,
Stuttgart.
Ullrich, C. J. (2009), ‘Realized volatility and price spikes in electricity markets:

The importance of observation frequency’, Working Paper, Virginia Polytechnic
Institute and State University, June 2009 .
van Groenendaal, W. J. H. (1998), ‘Estimating net present value variability for

deterministic models’, European Journal of Operational Research 107, 202–213.
van Groenendaal, W. J. & Kleijnen, J. P. (2002), ‘Deterministic versus stochastic

sensitivity analysis in investment problems: An environmental case study’, Euro-
pean Journal of Operations Research 141, 8–20.
van Treslong, A. & Huisman, R. (2009), ‘A comment on: Storage and the electricity
forward premium’, Research Paper, Erasmus Universiteit Rotterdam .
Vasicek, O. (1977), ‘An equilibrium characterization of the term structure’, Journal

of Financial Economics 5, 177–188.
Viehmann, J. (2009), ‘Risk premiums in the German day-ahead electricity market’,

EWI Working Paper No. 09.01, Universität zu Köln, June 2009 .
Vilanova, L. (2006), ‘Financial distress, lender passivity and project finance: The
case of Eurotunnel’, Working Paper, University of Lyon 2, September 2006 .
von Hirschhausen, C., Weigt, H. & Zachmann, G. (2007), ‘Preisbildung und Markt-
macht auf den Elektrizitätsmärkten in Deutschland’, Survey for Verband der In-
dustriellen Energie- und Kraftwirtschaft e.V., Technische Universität Dresden .
Vose, V. (2000), Risk Analysis: A Quantitative Guide, 2nd edn, John Wiley, New
York.
Wagner, U., Geiger, B., Hardi, M., Brückl, O., Roth, H. & Tzscheutschler, P.
(2004), CO2-Vermeidungskosten im Kraftwerksbereich, bei den erneuerbaren En-
ergien sowie bei nachfrageseitigen Energieeffizienzmassnahmen. Lehrstuhl für En-
ergiewirtschaft und Anwendungstechnik, Technische Universität München, April
2004.
Wawer, T. (2007), Förderung erneuerbarer Energien im liberalisierten deutschen

Markt, PhD thesis, Westfälische Wilhelms-Universität Münster.
223
Weber, C. (2010), ‘Adequate intraday market design to enable the integration of

wind energy into the European power systems’, Energy Policy 38, 3155–3163.
Weber, F., Schmid, T., Pietz, M. & Kaserer, C. (2010), ‘Simulation-based valua-
tion of project finance - Does model complexity really matter?’, Working Paper,
Technische Universität München, April 2010 .
Wei, S. Z. C. & Zhu, Z. (2006), ‘Commodity convenience yield and risk premium de-
termination: The case of the U.S. natural gas market’, Energy Economics 28, 523–
534.
Weigt, H. (2009), ‘A review of liberalization and modeling of electricity markets’,

Working Paper 34, Technische Universität Dresden, October 2009 .
Weigt, H. & von Hirschhausen, C. (2008), ‘Price formation and market power in the
German wholesale electricity market in 2006’, Energy Policy 36, 4227–4234.
Weron, R. (2006), Modeling and Forecasting Electricity Loads and Prices: A Statis-
tical Approach, 1st edn, John Wiley, New York.
Weron, R. (2008), ‘Market price of risk implied by Asian-style electricity options and
futures’, Energy Economics 30, 1098–1115.
Weron, R., Bierbrauer, M. & Trück, S. (2004), ‘Modeling electricity prices: Jump
diffusion and regime switching’, Physica A 336, 39–48.
Werthschulte, H. (2004), Kreditrisikomessung bei Projektfinanzierungen durch

Risikosimulationen, PhD thesis, Universität Münster.
Wilkens, S. & Wimschulte, J. (2007), ‘The pricing of electricity futures: Evidence

from the European energy exchange’, Journal of Futures Markets 27, 387–410.
Wimschulte, J. (2010), ‘The futures and forward price differential in the Nordic
electricity market’, Energy Policy 38, 4731–4733.
Working, H. (1949), ‘The theory of price of storage’, American Economic Review

39, 1254–1262.
Zachmann, G. & von Hirschhausen, C. (2008), ‘First evidence of asymmetric cost

pass-through of EU emissions allowances: Examining wholesale electricity prices
in Germany’, Economic Letters 99, 465–469.

Simulation-Based Valuation of Project Finance Investments - Crucial Aspects of Power Plant Projects

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Simulation-Based Valuation of Project Finance Investments - Crucial Aspects of Power Plant Projects

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Simulation-Based Valuation of Project Finance Investments - Crucial Aspects of Power Plant Projects

Uploaded by

Copyright:

Available Formats

TECHNISCHE UNIVERSITÄT MÜNCHEN

Lehrstuhl für Finanzmanagement und Kapitalmärkte

Simulation-based Valuation of Project Finance

Doktors der Wirtschaftswissenschaften

Vorsitzender: Univ.-Prof. Dr. Christoph Ann

Prüfer der Dissertation: 1. Univ.-Prof. Dr. Christoph Kaserer

2. Univ.-Prof. Dr. Gunther Friedl

4 Price Formation in the German Electricity Wholesale Market – An

5 The Impact of Model Complexity on the Simulation Results 178

6 Summary and Conclusion 193

A Hourly Prices Intraday Market 197

B Hourly Prices Day-Ahead Market 201

2.1 Loan-based Financing of Projects versus Project Finance . . . . . . . . 11

3.1 Schematic Functionality Project Finance Valuation Tool . . . . . . . . 97

4.1 Traded Volume Spot Market 2003-2009 . . . . . . . . . . . . . . . . . 131

5.1 Dimensions of Model Complexity . . . . . . . . . . . . . . . . . . . . . 180

A.1 Time Series Intraday Market, Hour Contract 1 - 4 . . . . . . . . . . . 197

B.1 Time Series Intraday Market, Hour Contract 1 - 4 . . . . . . . . . . . 201

3.1 Summary Input Parameters . . . . . . . . . . . . . . . . . . . . . . . . 101

4.1 Summary Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

APT Arbitrage Pricing Theory

ARCH Autoregressive Conditional Heteroskedasticity

ARMA Autoregressive Moving Average

BaFin Bundesanstalt für Finanzdienstleistungsaufsicht

BMWi Bundesministerium für Wirtschaft und Technologie

CAPM Capital Asset Pricing Model

CCC Constant Conditional Correlation

CF Cash Flow from Operations

CfD Contract for Difference

CFTC Commodity Futures Trading Commission

CO2 Carbon Dioxide

DCC Dynamic Conditional Correlation

DCF Discounted Cash Flow

DRA Debt Repayment Amount

DSCR Debt Service Cover Ratio

EAU Emission Allowance Unit

EBIT Earnings before Interest and Taxes

EBITDA Earnings before Interest, Taxes, Depreciation, and Amortization

ECC European Commodity Clearing

ECDP Expected Cumulative Default Probability

EEG Erneuerbare Energien Gesetz

EEX European Energy Exchange

EMEA Europe, Middle East and Africa

ENTSOE European Network of Transmission System Operators for Elec-

EU ETS European Union Emission Trading System

EUREX European Exchange

EXAA Energy Exchange Austria

FCFE Free Cash Flow to Equity

FCFF Free Cash Flow to the Firm

FFO Funds from Operations

FOCF Free Operating Cash Flow

GARCH General Autoregressive Conditional Heteroskedasticity

GDP Gross Domestic Product

GJR-GARCH Glosten, Jagannathan and Runkle - General Autoregressive Con-

IEA International Energy Agency

IFRS International Financial Reporting Standards

IGSSE International Graduate School of Science and Engineering

IRR Internal Rate of Return