Title: New Directions and Products in Energy Markets: Oil, Electricity, Carbon
1New Directions and Products in Energy Markets
Oil, Electricity, Carbon
- Hélyette Geman
- Professor of Finance
- Birkbeck, University of London ESSEC Business
School - To be presented at the Europlace Finance
Conference - - Paris - June 22, 2006
2Growth of 100 1991-1999
3Growth of 100 2000-2004
4(No Transcript)
5Is Mean-Reversion Dead ?
6Mean-Reversion versus "Random Walk"Representation
in Oil and Natural Gas Prices
- The objective is to check whether, in the
representation - the coefficient r is significantly different
from 1 - The H0 hypothesis is the existence of a unit root
(i.e., r 1) - A p-value smaller than 0.05 allows one to reject
the H0 hypothesis with a confidence level higher
than 0.95, in which case the process is of the
mean-reverting type - Otherwise, a unit root is uncovered and the
process is of the "random walk" type - The higher the p-value, the more the random walk
model is validated
7Mean-Reversion Properties for Oil and Natural
Gas Prices(Geman (2005) J. of Alternative
Investments)
- For crude oil
- a mean-reversion pattern prevails over the period
1994-2000 - it changes into a random walk (arithmetic
Brownian motion) as of 2000 - For natural gas
- there is a mean-reversion pattern until 2001
- since 2002, a change into a random walk accurs
- during both periods, seasonality of gas prices
tends to blur the signals
8US Natural Gas Prices over the periodJanuary
1994 - October 2004
- Spot prices are proxied by the Nymex one month
Futures contract - Over the period Jan 94 - Oct 04
- ADF p-value 0.712
- Phillips Perron p-value 0.1402
- Over the period Jan 1999 - Oct 2004
- ADF p-value 0.3567
- Phillips Perron p-value 0.3899
9- Taking instead log-prices, we obtain
- Over the last five years of the period, the
arithmetic Brownian motion assumption clearly
prevails
10WTI Spot Prices over the period January 1994 -
October 2004
- Again, spot prices are proxied by Nymex one-month
Futures prices - Whole period 1994-2004
- Augmented Dickey Fuller 0.651
- Phillips Perron 0.5048
- Period January 1999 - October 2004
- Augmented Dickey Fuller 0.7196
- Phillips Perron 0.5641
- The mean-reversion assumption is strongly
rejected over the whole period and even more so
over the recent one - Because of absence of seasonality, the property
is more pronounced than in the case of natural gas
11The Literature on Mean-Reversion in Commodity
Prices
- Bessembinder, Coughenour, Seguin and Smoller
(JOF, 1995) test the term structure of Futures
prices over the period January 1982 to December
1991 and find mean-reversion in the 11 markets
they examine. They also conclude that the
magnitude of mean-reversion is large for
agricultural commodities and crude oil, and
substantially less for metals. - Rather than examining evidence of ex post
reversion using time series of asset prices, they
use price data from futures contracts with
various horizons to test whether investors expect
prices to revert. The authors analyze the
relation between price levels and the slope of
the futures term structure an inverse relation
between prices and this slope constitutes
evidence that investors expect mean reversion in
spot prices, as it implies a lower rate of
expected intertemporal price appreciation when
prices rise
12- Pindyck (1999) analyses 127 years of data (period
1870-1996) on crude oil and bituminous coal,
obtained from the US Department of Commerce - Using a unit root test, he exhibits that prices
mean revert to stochastically fluctuating trend
lines these lines, which represent long-run
total marginal costs, are themselves unobservable - Pindyck finds that during the time period of
analysis, the random walk distribution for
log-prices, i.e., the geometric Brownian motion
for spot prices, is a much better approximation
for coal and gas than oil - The recent period (2000-2006) has been quite
different!
13(No Transcript)
14Other Relevant Models forElectricity Price
Processes
- A three-factor mean reverting process with
stochastic volatility -
- where the three Brownian motions may be
correlated
15Using Pure Jump Lévy Processes for Commodity
Prices
- Prices moves are represented as a succession of
jumps many small jumps, a few big jumps upward
or downward - The Lévy density k(x) translates the arrival
intensity of jumps of size x in a unit time
interval - The CGMY process
- (Carr-Geman-Madan-Yor, Journal of Business 2002)
- It is successfully implemented in many financial
institutions. Its Lévy density k is defined by
16Modelling the Electricity Price Processthe
Markov property
- One way of generating the downward part of a
spike is to introduce a non-Markovian process
the spot price needs to "remember" its past
values as well as the current value to know that
if should be going down (with a high probability) - Nearly all option pricing models in finance
involve Markovian processes - Another way of handling the representation of the
spike is to introduce in the dynamics of the
price process a component which triggers a
downward jump (in probability) only on the basis
of the current spot price and preserve the Markov
property!
17Checking that a Model forPower Spot Price is
Acceptable
- Realize that going from 4500 to 70 is
definitely a jump downward. - Hence, the model should allow for positive and
negative jumps - Generate with the model a variety of trajectories
and check that a least some of them look like
real trajectories - Trajectorial Adequacy of the model
- Compute the first 4 moments of the calibrated
model and of the real trajectory and verify that
they are similar - Statistical Adequacy of the model
- For instance, introducing only upward jumps
generates a very highly positive skewness which
is not observed in practice
18(No Transcript)
19(No Transcript)
20A Jump-Reversion ModelGeman - Roncoroni, J. of
Business (2006)
- A/ Empirical observations
- Trajectories Descriptive statistics
- Class of models to account for heterogeneity
across market - B/ Modeling by marked point processes
- C/ Calibration
- "Structural" elements
- Parameters
- Path Properties
- Statistics
21MAIN
-
Logarithmic
Prices
10
8
e
c
i
r
6
P
-
g
o
4
l
2
Years
22A Power Plant has the Option to transform Fuel
into Power
23Valuation of Physical Assets in the Energy
Industry
- Scenario A
- Gas 3.5 /MMbtu
- Power 40 /MWh
- Heat Rate 10 MMbtu/MWh
- Profit 40 - 3.5 x 10 5
- Scenario b
- Gas 4.5 /MMbtu
- Do not operate
- The ownership of the physical assets amounts to
a series of options over the lifetime of the
plant
24Pricing the Exchange Option
- It is the option to exchange at time T one asset
S2 for another asset S1 - Pay-off at time T max (0, S1(T) S2(T))
- the right numéraire is S2
- the relevant volatility is the volatility S of
- Margrabe's formula holds under the sole
hypothesis of a deterministic volatility for
no assumption on interest rates, stochastic or
not -
- where
-
-
- and
- The correlation coefficient plays a key role in
the option price
25Kyoto-Pricing Power Plant Valuation and Emission
Rights
- Spark Spread Pe (Pt/e) (PCO2 / e / c)
- where
- Pe Price of power in /MWh
- Pf Price of fuel
- E Efficiency
- PCO2 Price of Permits in /ton
- c Specific carbon content (ton CO2 / ton)
26- Kyoto pricing of a coal plant
27The European Carbon Market
- The 2005 emissions of the 21 States representing
88 of the European allowances were announced on
May 15, with a surplus over 3.4. The allowances
allocated to these countries were 62.8 Mt higher
than the declared emissions - The allowance price reacted very strongly to the
anticipated disclosure of compliance data and
dropped by 65 between April 24 and May 12, and
the price of the 2008 contract fell by 25 - The disclosure of compliance data did not give
the market enough information to limit its
instability price volatility in late May was
significantly greater than the one observed in
the 8 previous month - It seems necessary to improve the information in
direction of the market, by increasing
transparency and harmonization between countries
in their reporting