the systemic risk of energy markets - fgv epge · the systemic risk of energy markets diane pierret...

The Systemic Risk of Energy Markets

Diane Pierret

Université Catholique de Louvain & NYU

The Economics and Econometrics of Commodity Prices,

Rio de Janeiro 2012

Systemic Risk

Systemic risk: the risk of the financial sector as a whole being in distress

and its spillover to the economy at large

Systemic risk is related to

1 dependence, causality, externality, interconnectedness, spillover effects

2 market integration, contagion, commonality

3 shocks, crises, extreme events

Does this exist in energy markets?

How to measure it?

How does this affect the rest of the economy?

1 / 27

Energy Systemic Risk

Energy Systemic Risk: the risk of an energy crisis raising the prices of all

energy commodities with negative consequences for the real economy.

increased dependence of the economy on energy

increased energy market integration

extreme energy market shock from the supply side

negative consequences for the energy sector and the rest of the

economy

Energy Security: ”the uninterrupted availability of energy sources at an

affordable price.” (IEA)

2 / 27

Related literature

This project relates to the literature on

Past energy crises and the impact of energy prices on the economy

(Hamilton, 1983)

Systemic Risk

Network (Nier et al. (2007), Battiston et al. (2009), Billio et al.

(2010), Hautsch et al. (2011), Diebold and Yilmaz (2011))

Co-movements (Billio et al. (2010), Kritzman et al. (2011), Acharya et

al. (2010), Brownlees and Engle (2011))

Energy prices co-movements

Cointegration and causality in the mean (Bunn and Fezzi (2008))

Multivariate Volatility (Chevallier (2012), Bauwens et al. (2012))

Copulae (Benth and Kettler (2010), Boerger et al. (2009), Gronwald et

al. (2011))

3 / 27

Outline

1 The Energy Systemic Risk Measure: EnSysRISK

2 Econometric methodology and application

Causility in means and variances

Factor model and tail expectations

3 EnSysRISK and the impact on the economy

4 / 27

Outline






5 / 27

EnSysRISK

The conditional MES of an energy asset (Acharya et al. (2010), Brownlees and

Engle (2011)) is given by

MESit = Et−1 (rit |energy crisis)

Corresponding systemic prices are derived from past price levels

syspriceit = pit−1 ∗ exp(MESit)

EnSysRISK: the total cost of an energy asset to the non-energy sector during an

energy crisis

EnSysRISKit = max(0,syspriceit ∗wit),

where wit is the quantity exposure of the economy to asset i at time t.

For an energy contract with maturity and delivery period ν , the exposure at time

t is

wit(ν) = ςiEt−1 (finconsτ − inv τ) for τ ∈ [t +ν ; t +2ν−1]

6 / 27

The MES of energy markets (1/2)

Energy crisis: an extreme positive energy market shock from the supply side

MESit(C ) = Et−1 (rit |rEnM,t > C , rM,t < 0) ,

where rEnM,t is the energy market return, rM,t is the return of the non-energy

sector, and C represents the VaR of the energy market at (1−α)%.

An extreme increase in energy prices...

Not only oil prices: “While the security of oil supplies remains important,

contemporary energy security policies must address all energy sources” (IEA

2011)

Integration dimensions: underlying energy (oil, coal, natural gas, electricity,

carbon), maturity/delivery, region

Futures prices

7 / 27

Phelix Baseload index = average of 24 day-ahead spot prices

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

100

200

300

400

500

EEX - Phelix Base Hr.01-24 E/Mwh

EEX - Phelix Peak Hr.09-20 E/Mwh

Futures are “insurance contracts” providing protection against the price

uncertainty in the spot market

!"#$%$ &'(' Spot and future trading in the EEX

!"#$%&'(%)*

!"#$%(%)*

)*+,-./0/,*121#3

%4/,561$74089/6:

;4+<

+,-./%%0)*1

&%2 34%5

=/+,0*,-./0/,*1>1#3

#4?0@-./0/,*1(1#3

)*+,-./0/,*121#3

#4?0@-./0/,*1A1#3

%4/,561$74018B*55:

Source: University St Gallen

The MES of energy markets (2/2)

Methodology: the conditional MES as a function of mean, volatility and tail

expectation

MESit(C ) = Et−1 (µit +σituit |rEnMt > C , rMt < 0)

= µit +σitEt−1 (uit |rEnMt > C , rMt < 0)

where µit and σ2it are the conditional mean and variance of asset return i and

uit = (rit −µit)/σit are the standardized residuals.

Separate causality from common factor exposure:

Causality in µit and σit

Heteroskedasticity and causality are removed to concentrate on the ’pure’

commonality or contagion phenomenon (Forbes and Rigobon (2002), Billio

and Caporin (2010))

Common factors in standardized residuals: uit = f (yt ,ζit)

8 / 27

Outline






9 / 27

EEX futures, energy spot and DAX industrial indices

10 EEX futures

Electricity: Phelix financial futures (M, Q, Y maturity)

Natural gas: Gaspool physical futures (M, Q, Y maturity)

Coal: ARA financial futures (M, Q, Y maturity)

EU emission allowances: EUA financial futures (Y maturity)

3 energy spot indices highly correlated to EEX futures returns

Brent crude oil

European coal

EUA

1 DAX industrial index (energy consumers) = non-energy index

Market areas

NetConnect Germany (NCG):

· Open Grid Europe GmbH

· Bayernets GmbH

· Eni Gas Transport Deutschland

· GRTgaz Deutschland GmbH

· GVS Netz GmbH

· Thyssengas GmbH (planned in 2nd quarter 2011)

GASPOOL:

· Gasunie Deutschland

· Ontras – VNG Gastransport GmbH

· Wingas Transport GmbH & Co. KG

Title Transfer Facility (TTF): (planned in 1st quarter 2011)

· Gas Transport Services B. V.

Active trading participants on the EEX Natural Gas Market

22 c o n n e c t i n g m a r k e t s

Natural Gas

Market Areas

EEX offers trading in natural gas on the Spot Market for the NetConnect Germany, GASPOOL and TTF (planned for 1st quarter 2011) market areas. On the Derivatives Market gas trading transactions can be concluded in the NCG and GASPOOL market areas.

Spot Market Derivatives Market

Trading Participants

EEX has been recording a continued positive trend as regards the number of trading participants: At the end of the year 2010 the number of trading participants in gas trading had increased to a total of 97 comparing with 76 companies at the beginning of the year. Thus, EEX is the gas exchange in Continental Europe with the highest number of trading participants.

60

50

40

30

20

10

01/2010 2/2010 3/2010 4/2010

3639

4648

26 26 2722

Source: EEX 10 / 27

Outline






11 / 27

Price series

MPhelix (EUR/MWh)

2008 2010 2012

25

50

75

100

Source: Datastream

MPhelix (EUR/MWh) QPhelix (EUR/MWh)

2008 2010 2012

25

50

75

100 QPhelix (EUR/MWh) YPhelix (EUR/MWh)

2008 2010 2012

25

50

75

100 YPhelix (EUR/MWh) MGaspool (EUR/MWh)

2008 2010 2012

10

20

30

MGaspool (EUR/MWh)

QGaspool (EUR/MWh)

2008 2010 2012

20

30

40 QGaspool (EUR/MWh) YGaspool (EUR/MWh)

2008 2010 2012

20

30

40YGaspool (EUR/MWh) MARA (EUR/ton)

2008 2010 2012

50

100

150 MARA (EUR/ton) QARA (EUR/ton)

2008 2010 2012

50

100

150 QARA (EUR/ton)

YARA (EUR/ton)

2008 2010 2012

75

100

125

150 YARA (EUR/ton) YEUA (EUR/ton CO2e)

2008 2010 2012

10

20

30

YEUA (EUR/ton CO2e) MLCX EUA spot (EUR/ton CO2e)

2008 2010 2012

100

150

200

MLCX EUA spot (EUR/ton CO2e) Brent crude oil (EUR/barrel)

2008 2010 2012

25

50

75

100 Brent crude oil (EUR/barrel)

MLCX European coal spot (EUR/ton)

2008 2010 2012

100

200

300

MLCX European coal spot (EUR/ton) DAX Industrial (EUR)

2008 2010 2012

2000

3000

4000DAX Industrial (EUR)

12 / 27

Cointegration and Granger-causality in the mean

Augmented Vector Error Correction Model (VECM) for the joint mean of the

(n×T ) matrix of returns rt capturing cointegration, autocorrelation, causality,

and seasonality

rit = πiη �ln(pt−1)+

K

∑k=1

δ �ik rt−k +M

∑m=1

θ �imxt−m +ϕ �i qt + εit

where η are the cointegrating vectors,πi are error-correction parameters,δik is a (n×1) vector of autocorrelation and Granger-causal parameters of order k,xt−m are exogenous variables lagged by m days andqt are deterministic (seasonal) factors.

Similar to Bunn and Fezzi (2008), except that all energy products are here

considered to be endogenous variables (as part of the ’system’).

13 / 27

Multiplicative Causality GARCH modelMultiplicative Causality GARCH model for the variance with a GARCH

component and an interaction component

εit = σituit =�

φitgituit

where

git = (1−αii −βi −γii

2)+αii

� ε2it−1

φit−1

�+βigit−1 + γii

� ε2it−1

φit−1

�I{εit−1<0},

φit = f (u1t−1, ...,ui−1,t−1,ui+1,t−1, ...,unt−1) li (t),

I{εit−1<0} is a dummy variable equal to one when the past shock of asset i is

negative, and li (t) is a deterministic function of time.

In this application:

φit = ci exp

�n

∑j=1,j �=i

(ϑijujt−1 +αij |ujt−1|)+κ �i dt

�

where dt are deterministic terms including seasonal dummies.14 / 27

Networks of Causal Relationships

Causal relationships reflect physical relationships in the energy market

(subsitution, merit-order) and spillover effects

Mean Network Variance Network

15 / 27

Multiplicative Causality GARCH VariancesLog-likelihood GARCH = -27335

Log-likelihood MC-GARCH = -27059

2008 2009 2010 2011

10

20

30

MC-GARCH Electricity GARCH Electricity

2008 2009 2010 2011

10

20

MC-GARCH Natural Gas GARCH Natural Gas

2008 2009 2010 2011

10

20

MC-GARCH Coal GARCH Coal

2008 2009 2010 2011

10

20

30 MC-GARCH Carbon GARCH Carbon

2008 2009 2010 2011

10

20

30

MC-GARCH Brent GARCH Brent

2008 2009 2010 2011

25

50

MC-GARCH DAX GARCH DAX

16 / 27

Outline






17 / 27

Factor model: Dynamic PCA

Dynamic PCA based on the daily correlation matrices estimated with the

Dynamic Conditional Correlation (DCC) model

Ht = DtRtDt = Dt�AtΛtA

�t +Rζt

�Dt

where Dt = diag(σ1t , ...,σnt), At is a matrix of s eigenvectors associated with the s largesteigenvalues that are contained in the diagonal matrix Λt = diag(λ1t ,λ2t , ...,λst) withλ1t ≥ λ2t ≥ ...≥ λst , s ≤ n and Rζt is the correlation matrix of idiosyncratic terms ζt .

Standardized residuals uit = (rit −µit)/σit becomes a function of the first s

dynamic principal components and idiosyncratic terms

uit =s

∑j=1

aijtyjt +ζit

where aijt is the element of the eigenvector associated with asset i and principal

component yjt extracted from Rt , and ζit = uit −∑sj=1 aijtyjt .

18 / 27

Restrictions on the dynamic PCAThe energy crisis condition is defined by 2 factors:

Et−1 (uit |energy crisis) := Et−1 (uit |rEnMt > C , rMt < 0)

the return on the non-energy sector: rMt � yMt = y1t = a�1tut

the energy market return: rEnMt � yEnMt = y2t = a�2tut

Restricted PCA (Ng et al. (1992)): the 1st component is restricted to be the

non-energy return (DAX industrial index)

maxa1t a�1tRta1t

s.t. a�1ta1t = 1, ai1t = 0 ∀t,∀i �= DAX industrial

The other dynamic components are mutually orthogonal and orthogonal to the

industrial component

maxajt a�jtRtajt

s.t. a�jtajt = 1, a

�jtalt = 0 ∀t,∀l �= j

19 / 27

Energy Market Integration

The contribution of the energy trend component (λEnM,t/∑nj=1 λj ,t) as a

dynamic measure of energy market integration (Billio et al. (2010),

Kritzman et al. (2011))

2009 2010 2011

40.0

42.5

45.0

47.5

50.0

52.5

55.0

57.5

60.0

LB b

ankr

uptc

y

Fuku

shim

a pl

ant o

utag

e

BP’

s rig

exp

losi

on

Rus

sia-

Ukr

aine

NG

dis

pute

Hig

hest

cru

de o

il pr

ice

Energy trend component contribution (%)

Decreasing crude oil prices

20 / 27

Energy Systematic Risk

The correlation with the energy trend component (λEnMta2i ,EnMt) as a

measure of energy systematic risk

2009 2010 2011

0.6

0.7

0.8 Cor(Electricity, Energy Market)

2009 2010 2011

0.6

0.7

0.8 Cor(Natural Gas, Energy Market)

2009 2010 20110.7

0.8

0.9 Cor(Coal, Energy Market)

2009 2010 2011

0.4

0.6

0.8 Cor(Carbon, Energy Market)

2009 2010 2011

0.2

0.4

0.6

Cor(Brent, Energy Market)

21 / 27

Tail expectations

The non-energy market return: rMt � yMt = y1t = a�1tut

The energy market return: rEnMt � yEnMt = y2t = a�2tut

The tail expectation Et−1 (uit |energy crisis) is approximated by

s

∑j=1

[aijtEt−1 (yjt |yEnMt > C , yMt < 0)]+Et−1 (ζit |yEnMt > C , yMt < 0)

A nonparametric estimator of tail expectations is

E(yjt |yEnMt > C , yMt < 0) =∑T

τ=1 yjτΦ

��yEnMτ√

λEnMτ− C√

λEnMt

�h−1

�I (yMτ < 0)

∑Tτ=1 Φ

��yEnMτ√

λEnMτ− C√

λEnMt

�h−1

�I (yMτ < 0)

where Φ(·) is the Gaussian cummulative distribution function. The same

estimation procedure applies to E(ζit |yEnMt > C , yMt < 0).

22 / 27

The conditional MES of energy assets

MESit(C ) = µit +σitEt−1

�s

∑j=1

aijtyjt +ζit |yEnMt > C , yMt < 0

�

2009 2010 2011

2.5

5.0

7.5MES Electricity

2009 2010 2011

2.5

5.0

MES Natural Gas

2009 2010 2011

2.5

5.0

7.5

MES Coal

2009 2010 2011

2.5

5.0

7.5 MES Carbon

2009 2010 2011

0.0

2.5

5.0 MES Brent

23 / 27

Outline






24 / 27

EnSysRISKEnSysRISK = the total cost in million euros of each energy commodity class to

the German non-energy sector during a potential energy crisis

2009 2010 2011

75

100

125

150 EnSysRISK Electricity

2009 2010 2011

20

40

60

EnSysRISK Natural Gas

2009 2010 2011

2

4

6

EnSysRISK Coal

2009 2010 2011

10

15

20

25 EnSysRISK Carbon

2009 2010 2011

50

100

150EnSysRISK Brent

25 / 27

’Net’ impact on the economyMESMt(C ) = µMt +σMtEt−1

�∑s

j=1 aMjtyjt +ζMt |yEnMt > C , yMt < 0�

’Net’ impact of the energy crisis:

∆MESMt(C ) = MESMt(C )−MESMt(VaR(rEnMt)0.5)

MES DAX .95 MES DAX .5

2009 2010 2011

-4

-2

0MES DAX .95 MES DAX .5

DeltaMES DAX

2009 2010 2011

-0.75

-0.50

-0.25

0.00 DeltaMES DAX

26 / 27

Summary

Energy Systemic Risk: the risk of an energy crisis raising the prices of all energy

commodities with negative consequences for the real economy

EnSysRISK: the total cost of an energy commodity to the rest of the economy

during an energy crisis

EnSysRISK increases with

high MES

high prices

high dependence of the economy on the energy source

The MES captures co-movements in energy assets

Causal relationships in means and variances reflect possible spillover effects

from one product to another

Tail exposure to common factors: the MES is conditional on extreme energy

market shocks from the supply side (restricted dynamic PCA)

Contact: [email protected]

27 / 27

the systemic risk of energy markets - fgv epge · the systemic risk of energy markets diane pierret...

Documents