Measuring Performance of Long-Term Power Generating Portfolios - Green Energy and Efficiency: An Economic Perspective - page 389

quantify the value of a number of options that (project) managers have at their
disposal (e.g. the investment timing, size, stages, and so on). See Dixit and Pindyck
[ 14 ] and Trigeorgis [ 34 ].
When it comes to applying ROA to inform investments in power technologies, it
is usually necessary to adopt relatively restrictive assumptions about the stochastic
behavior of commodity prices. Besides, futures contracts on those commodities
may well be available but their liquidity for the decades-long maturities that these
infrastructures typically involve may falter. For a sample of ROA applications see
Murto and Nese [ 26 ], Roques et al. [ 31 ], N

ä ss ä k ä l ä and Fleten [ 27 ], Blyth et al. [ 9 ],

Abadie and Chamorro [ 1 ].
Investors in liberalized electricity markets are naturally concerned about the
expected return and the risk of their investments. At the same time, policy makers
may guide investments in power plants in a particular direction (e.g. by adopting a
societal, as opposed to private, perspective). We propose a model for assessing the
performance of dynamic generation mixes in a mean-variance context. In particular,
we focus on the expected price of electricity and the price volatility that result from
different generating portfolios that change over time (because of new investments
and decommissioning of old plants).
There is a stark difference between our approach and the MVP portfolio
approach. The latter typically aims to identify a set of ef cient fuel mixes that

optimally trade off the risks and expected returns of diversied portfolios of gen-
erating plants. This

, however, usually corresponds to a single-
period uncertain situation, i.e. adopts a static perspective. Instead, we develop a
dynamic, multi-period approach. We assess the performance of different generating
mixes over decades. Similarly to the mean-variance approach, we can restrict
ourselves to considering a handful of particular generation settings which are of
interest to industry or policy makers. Our two measures can be plotted in the
standard risk-expected cost (or return) space, just like in the portfolio approach. But
they tell a rather different story, namely how our time-varying

' ef cient frontier '

' portfolios ' behave

over a multi-year period (in terms of electricity price).
The model comprises two stages, namely simulation and optimization. The
optimization model minimizes an objective function subject to constraints. The
objective function considers two kinds of system costs: those of electricity gener-
ation and of unserved or lost load. The constraints can be split into two blocks
concerning the physical and economic environment. Regarding physical uncer-
tainty, power infrastructures are subject to failure. As for economic uncertainty,
commodity prices display mean reversion and seasonality where appropriate. Load
is similarly assumed to be seasonal and stochastic. The optimization provides, at
any time, the level of generation from each technology and served load along with
aggregate generation costs, carbon emissions, and allowance costs. We consider a
20-year time horizon (the one adopted in the UK Future Energy Scenarios). Over
this period the network topology changes naturally as new stations start operation