When regulation brings its own host of distortions and
inefficiencies, the mere existence of a market failure is
insufficient to ensure that government intervention will improve
welfare. Instead, by comparing the distortions under potential
regulatory regimes, one can identify superior policies as those
with relatively fewer imperfections. My research undertakes such an
evaluation in the context of U.S. wholesale electricity markets,
which have replaced command-and-control-type operations in some
areas.
To do so, I construct a virtually complete hourly
characterization of supply and demand of the U.S. electrical grid
from 1999-2012. Data on fuel costs, capacities, heat efficiency,
and operations of nearly all generating units at the hourly level
allow me to construct power supply curves (known as the
“merit order”) for each of 98 “Power Control
Areas” (PCAs), as well as observe the units that were chosen
to operate to meet demand at any moment in time. These curves allow
me to calculate two key welfare measures for each PCA-date-hour:
“out of merit” losses from dispatching higher marginal
cost units relative to installed capacity, and the gains from
trading electricity across areas. Market power losses manifest
themselves as out-of-merit production, as do normal grid operations
such as maintenance, refueling, start-up costs, and transmission
congestion. In either case, the increased operational costs are
observationally equivalent as the distance between the realized
cost of operations and cost from utilizing only the lowest-cost
installed capacity.
Continue to full version
I develop a framework and compile the necessary data to examine
both gains from trade and out-of-merit losses over the history of
market transitions since 1999. I use the staggered creation and
expansions of wholesale electricity markets over this period to
estimate the causal impact of using markets to allocate production
on these welfare measures. I estimate changes in gains from trade
and out-of-merit losses following the transition to market dispatch
against PCAs that have not undergone any regulatory changes. This
approach finds that gains from trade increase by upwards of 30
percent after adopting market dispatch due to a 10 percent increase
in electricity traded. There is also a 10 percent decrease in
out-of-merit operations, reducing these costs by nearly 20
percent.
This simple approach is susceptible to the confounding effects
of fuel price fluctuations (over time and across areas) when
estimating counterfactual outcomes: fuel prices shift supply
curves, making historical outcomes poor counterfactuals for what
would have happened 2 today under a different set of prevailing
fuel prices. This means one might estimate changes in the gains
from trade without any actual changes in production patterns
because the value of offset production scales with fuel prices.
This issue motivates a “policy function” approach in
which I estimate each system operators’ rules for dispatching
units in a given year, and compare outcomes the following year
against those predicted by the policy function. I show how the
treatment effect can be estimated by comparing changes in the
quality of fit of this rule across areas that switch to market
dispatch against areas with no change in regulation.
Estimating dispatch probabilities with out-of-sample validity is
a pure prediction problem for which recent developments in the
machine learning literature have proven to be particularly
effective. I use such methods to estimate the policy functions,
then embed the results in a framework to estimate causal treatment
effects.
This approach yields estimates smaller in magnitude than the
simpler estimates for gains from trade, suggesting fuel price
confounding. I find that production costs are reduced by about
three billion dollars per year due to market-based improvements in
allocating output to lower cost units, with these savings split
between reduced output from uneconomical units and gains from trade
by a factor of 2 to 1.
I note that my estimates measure changes in how output is
allocated, given the installed capacity, costs, and patterns of
demand. It would not be unreasonable to suspect that market
dispatch has affected investment incentives, which are likely an
important source of welfare changes. In addition, my estimates
measure the average effect of market dispatch, which itself has
been heterogeneous both with respect to preexisting institutions
(i.e., power pools, bilateral markets, or smoke-filled rooms), and
with respect to the rules of the markets implemented (uniform or
locational marginal prices, virtual bidding, market monitors,
etc.).
However, given the even greater differences between market and
traditional dispatch methods, these estimates should be informative
regarding the performance of the relatively new mechanisms that
currently determine how more than 60 percent of generating capacity
in the United States is utilized.
NOTE:
This research brief is based on Steve Cicala, “Imperfect
Markets versus Imperfect Regulation in U.S. Electricity
Generation,” January 22, 2017, https://epic.uchicago.edu/research/
publications/imperfect-markets-versus-imperfect-regulationus-electricity-generation.
