Measuring Performance of Long-Term Power Generating Portfolios - Green Energy and Efficiency: An Economic Perspective - page 398

estimate the underlying parameters of wind generation, pumped storage and hydro
generation; see Appendix Tables A.3 , A.5 , and A.7 .
Any day we have futures prices of all contracts on natural gas with monthly,
quarterly, seasonal (April

March), and yearly maturities
on the European Energy Exchange (EEX, Leipzig). We collected these data over
231 days. Similarly for coal to be delivered in Amsterdam, Rotterdam, or Antwerp
(so-called ARA coal). We also collected the prices of futures contracts on EU
emission allowances traded on the Intercontinental Exchange (ICE; London); see
Chamorro (2012), Appendix D. Using the futures prices on each day and non-
linear least-squares, we derived the curve that best

September and October

- -

ts futures prices on that day;
this provides an estimate of the parameters in the (risk-neutral) stochastic model.
Upon the calibration on each of the sample days, we computed the corresponding
average values in a second step; we use them as reasonable estimates of future
behavior.
Concerning the economic dispatch, the system operator aims to nd an optimal

vector of power generated that minimizes the sum of (bid-

based) generation costs and unserved demand costs subject to the restrictions stated
above. The number of possible states of the system is 2 ð 22 þ 79 þ 10 Þ in 2012; this

ðxÞ and consumed ðsÞ

gure will change as old plants are decommissioned and new plants start operation.
Our aim is to evaluate the performance of dynamic generation portfolios. We
discount future cash-

ows at the risk-free interest rate using risk-neutral parameters.
We run 750 simulations each consisting of 1,200 steps over 20 years (i.e. ve steps

per month). At each step the optimal dispatch problem is solved subject to the
restrictions then in place; i.e. we solve 900,000 optimization problems that mini-
mize the sum of the bid-based costs of electricity generation and the cost of
unserved load, subject to linear and non-linear restrictions. The solution to each
problem de nes the levels of generation and the power effectively served. Hence we

compute the bid-based production costs, electricity price, and carbon emissions,
among other variables. We follow the same steps with each generation portfolio.
The comparison among them describes their (relative) performance in terms of the
variable(s) involved.

3.1 Future Demand: Assumptions

We collected monthly load data from January 2002 to August 2013, i.e. 140
observations; see Fig. 1 . Our base case analysis assumes that electricity demand
shows mean reversion over time with a null rate of growth. Transmission and
distribution losses alongside theft account for 9 % of overall demand over the
sample period. We estimate a load function with seasonality; see Appendix
Table A.1 . The model is run with the same forecast demand under all the generation
mixes considered.