Model Risk and Power Plant Valuation Energy & Finance Conference 2013, Essen, 9-11 October 2013 Anna Nazarova Based on a joint work with Karl Bann ¨ or, R¨ udiger Kiesel and Matthias Scherer | Chair for Energy Trading and Finance | University of Duisburg-Essen
30
Embed
Model Risk and Power Plant Valuation - uni-due.de€¦ · Seite 3/30Model Risk and Power Plant Valuation j Motivation and Introduction Motivation and Introduction: Questions I How
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Model Risk and Power Plant Valuation
Energy & Finance Conference 2013, Essen,9-11 October 2013
Anna NazarovaBased on a joint work with Karl Bannor, Rudiger Kiesel and Matthias Scherer | Chair for Energy Trading and Finance |University of Duisburg-Essen
Seite 2/30 Model Risk and Power Plant Valuation |
OutlookMotivation and Introduction
Theoretical Aspects
Spread Options and Power Plant Valuation
Empirical Investigation of the Model Risk
Questions and discussion
References
Appendix
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 3/30 Model Risk and Power Plant Valuation | Motivation and Introduction
Motivation and Introduction: Questions
I How significant is the impact of the model’s choice on the valueof a given instrument?
I How to assess the value of the parameters’ uncertainty?
I What is the main driver of the model risk in the energy markets?
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 4/30 Model Risk and Power Plant Valuation | Motivation and Introduction
General Approach
I We consider the model risk inherent in the valuation procedure offossil power plants.
I We focus on a gas-fired power plant as flexible and low-carbonsource of electricity which is an important building block in termsof the switch to a low-carbon energy generation.
I We model the generated financial streams as the spread optionand investigate the reinvestment problem.
I To capture model risk we use a methodology recently establishedin a series of papers: [Cont, 2006, Bannor and Scherer, 2013].
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 5/30 Model Risk and Power Plant Valuation | Theoretical Aspects
Risk-Captured Price
Having
I a contingent claim X ,I a parameter space Θ,I a distribution R on the parameters,I a parameterised family of valuation measures (Qθ)θ∈Θ,I a law-invariant, normalised convex risk measure ρ
results in a risk-captured price of a contingent claim X by
Γ(X ) := ρ(θ 7→ Eθ[X ]).
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 6/30 Model Risk and Power Plant Valuation | Theoretical Aspects
Visualisation of the Steps of Parameter Risk-Capturing Valuation
Quantifies parameter risk of derivative price
Model: complex financial market
Discounted derivative payout X
Parameter space Θ Derivative price Eθ[X]
Probability measure R on Θ
Ask price: Г(X)= ρ(θ → Eθ[X])
Bid price: -Г(-X)
Risk measure ρ
Derivative price distribution
induced by R and θ → Eθ[X]
Pricing function θ → Eθ[X]
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 7/30 Model Risk and Power Plant Valuation | Theoretical Aspects
Risk-Capturing Functional: AVaR Example
I Define AVaR w.r.t. the significance level α ∈ (0,1) of somerandom variable X as
AVaRα(X ) =1α
∫ α
0qX (1− β)dβ,
I where qX (γ) is γ quantile of the random variable X .
I The AVaR measures the risk which may occur according to thepreviously specified model Qθ.
I When calculating the parameter risk-captured price of X beinginduced by the AVaR, risk-neutral prices (Eθ[X ])θ∈Θ w.r.t.different models (Qθ)θ∈Θ are compared and subsumed by theAVaR risk measure. Hence, the AVaR is used to quantify theparameter risk we are exposed to when pricing X .
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 8/30 Model Risk and Power Plant Valuation | Spread Options and Power Plant Valuation
Clean Spark Spread Option and Virtual Power Plant
I We model the daily profit (or loss) of the virtual power plantposition as
Vt = maxPt − hGt − ηEt ,0,
I Pt - is the power price;I Gt - is the gas price;I Et - is the carbon certificate price;I h - is the heat rate of the power plant;I η - CO2 emission rate of the power plant.
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 9/30 Model Risk and Power Plant Valuation | Spread Options and Power Plant Valuation
Energy Price Models
LetI (Ω,P,F ,Ft,t∈[0,T ]) be a complete filtered probability space;I carbon price
W E , W G and N are mutually independent processes,dW P
t dW Gt = ρdt .
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 10/30 Model Risk and Power Plant Valuation | Empirical Investigation of the Model Risk
Data
I Phelix day base: It is the average price of the hours 1 to 24 forelectricity traded on the spot market. It is calculated for allcalendar days of the year as the simple average of the auctionprices for the hours 1 to 24 in the market area Germany/Austria,EUR/MWh.
I NCG daily price: Delivery is possible at the virtual trading hub inthe market areas of NetConnect Germany GmbH & Co KG,EUR/MWh.
I Emission certificate daily price: One EU emission allowanceconfers the right to emit 1 tonne of carbon dioxide or 1 tonne ofcarbon dioxide equivalent, EUR/EUA.
I Observation period: 25.09.2009 - 08.06.2012.
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 11/30 Model Risk and Power Plant Valuation | Empirical Investigation of the Model Risk
Power, Gas, and Carbon Prices, 25.09.2009 - 08.06.2012
!"#$
!%#$
!&#$
&#$
%#$
"#$
'#$
(#$
)(*#"*)#
#($
)(*#+*)#
#($
)(*#'*)#
#($
)(*#,*)#
#($
)(*#(*)#
#($
)(*&#*)#
#($
)(*&&*)#
#($
)(*&)*)#
#($
)(*#&*)#
&#$
),*#)*)#
&#$
%&*#%*)#
&#$
%#*#-*)#
&#$
%&*#"*)#
&#$
%#*#+*)#
&#$
%&*#'*)#
&#$
%&*#,*)#
&#$
%#*#(*)#
&#$
%&*&#*)#
&#$
%#*&&*)#
&#$
%&*&)*)#
&#$
%&*#&*)#
&&$
),*#)*)#
&&$
%&*#%*)#
&&$
%#*#-*)#
&&$
%&*#"*)#
&&$
%#*#+*)#
&&$
%&*#'*)#
&&$
%&*#,*)#
&&$
%#*#(*)#
&&$
%&*&#*)#
&&$
%#*&&*)#
&&$
%&*&)*)#
&&$
%&*#&*)#
&)$
)(*#)*)#
&)$
%&*#%*)#
&)$
%#*#-*)#
&)$
%&*#"*)#
&)$
!"#$%&'(
)*%&
+(,%&
./0123$ 456$$ 5789:;$
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 12/30 Model Risk and Power Plant Valuation | Empirical Investigation of the Model Risk
Clean Spark Spread, 25.09.2009 - 08.06.2012
-‐60.00
-‐40.00
-‐20.00
0.00
20.00
40.00
60.00
29.05.20
09
29.06.20
09
29.07.20
09
29.08.20
09
29.09.20
09
29.10.20
09
29.11.20
09
29.12.20
09
29.01.20
10
28.02.20
10
31.03.20
10
30.04.20
10
31.05.20
10
30.06.20
10
31.07.20
10
31.08.20
10
30.09.20
10
31.10.20
10
30.11.20
10
31.12.20
10
31.01.20
11
28.02.20
11
31.03.20
11
30.04.20
11
31.05.20
11
30.06.20
11
31.07.20
11
31.08.20
11
30.09.20
11
31.10.20
11
30.11.20
11
31.12.20
11
31.01.20
12
29.02.20
12
31.03.20
12
30.04.20
12
31.05.20
12
Spark Spread
Value
Date
Spark Spread
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 13/30 Model Risk and Power Plant Valuation | Empirical Investigation of the Model Risk
Estimating the Model Parameters
I Estimation of the seasonal trends and deseasonalisation ofpower and gas.
I Separation of the power base and spike signals.
I Estimation of the mean-reverting rates.
I Estimation of the power base signal Xt .
I Estimation of the spike signal Yt .
I Estimation of the correlation.
Following the above steps, we estimate the set of parameters withmainly an MLE approach
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 14/30 Model Risk and Power Plant Valuation | Empirical Investigation of the Model Risk
General Procedure: Step 1
I After estimating all the parameters of our prices, we simulatethem for the future time period and compute for every day t thespark spread value Vt given as
Vt = maxPt − hGt − ηEt ,0.
I Then, by fixing all the parameters except for the chosen one andsetting the shift value ξ (e.g. 1%), we compute shifted up anddown spark spread values as
V upt (θ + ξ),
V downt (θ − ξ).
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 15/30 Model Risk and Power Plant Valuation | Empirical Investigation of the Model Risk
General Procedure: Step 2
I We compute the value of the power plant (VPP) by means ofMonte Carlo simulations. For fixed large N and T = 3 years wehave
VPP(t ,T ) =1N
N∑i=1
VPPi (t ,T ),
VPPi (t ,T ) =
∫ T
te−r(s−t)Vi (s) ds.
I For the chosen shift ξ we also compute
VPPup(t ,T ; θ) = VPP(t ,T ; θ + ξ),
VPPdown(t ,T ; θ) = VPP(t ,T ; θ − ξ).
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 16/30 Model Risk and Power Plant Valuation | Empirical Investigation of the Model Risk
General Procedure: Step 3
I We continue with sensitivity measuring of the VPP w.r.t. theparameter θ with the central finite difference [Glasserman, 2004]
∇θVPP(t ,T ) :=∂VPP∂θ
=VPPup(t ,T ; θ)− VPPdown(t ,T ; θ)
2ξ
I Finally, we compute the bid and ask prices by using aclosed-form approximation formula for the AVaR to get therisk-captured prices by subtracting and adding risk-adjustmentvalue to VPP(t ,T ) respectively. This risk-adjustment value iscomputed as
ϕ(Φ−1(1− α))
α
√(∇θVPP)′ · Σθ · ∇θVPP
N,
denoting by Σθ the asymptotic covariance matrix of the estimatorfor the parameter θ [McNeil et al., 2005].
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 17/30 Model Risk and Power Plant Valuation | Empirical Investigation of the Model Risk
Risk Values Results
I parameter risk in spike size: Laplace and Gaussian distributions;I parameter risk in correlation;I parameter risk in gas signal;I joint parameter risk in gas and base power signal;I joint parameter risk in gas, power and emissions (all processes
except of jump size parameter).
Anna Nazarova | Chair for Energy Trading and Finance, University of Duisburg-Essen |
Seite 18/30 Model Risk and Power Plant Valuation | Empirical Investigation of the Model Risk
Resulting values for the relative width of the bid-ask spread forvarious model risk sources. α1 = 0.01 (the highest risk-aversion),α2 = 0.1, α3= 0.5