Read Microsoft Word - executive summary.doc text version


prepared by London Economics International LLC

Julia Frayer, Amr Ibrahim, Serkan Bahçeci, Sanela Pecenkovic 717 Atlantic Avenue, Unit 1A Boston, MA 02111

January 31, 2007

About the Authors

Julia Frayer, as Managing Director, co-leads the firm's market analysis and quantitative business practice area, which involves economic analysis and evaluation of infrastructure assets, detailed modeling of electricity markets, and price forecasting. She has worked extensively in the power sector (covering all aspects of the value chain), as well as other infrastructure industries such as the natural gas sector and the water sector. On regulatory front, she has specialized in areas related to market power mitigation, auction design, and performance-based ratemaking. Dr. Amr Ibrahim is a Senior Consultant with LEI specializing in restructuring, market design, regulation, and operation of wholesale and retail energy markets in the US, Canada, and South America. Dr. Serkan Bahçeci is a Consultant with LEI specializing in empirical analysis and applied econometrics to electricity markets. He has worked extensively on engagements involving generation market power and strategic bidding. Sanela Pecenkovic is a consultant with LEI, providing research and analytical support to market analysis-related engagements in the power sector for markets across North America.


This study has been produced by London Economics International LLC ("LEI"), an independent consulting firm specializing in economic analysis for the infrastructure industries. The study was funded by American Public Power Association ("APPA") and the National Rural Electric Cooperative Association ("NRECA"). APPA and NRECA make no warranty, express or implied, and assume no legal liability for the information in this report; nor does any party represent that the uses of this information will not infringe upon privately owned rights.

Executive summary

In 2006, the American Public Power Association ("APPA") launched the Electric Market Reform Initiative ("EMRI") to investigate the challenges facing wholesale electricity markets. The first phase of this initiative entails a series of detailed studies of wholesale electricity markets in the United States. One of these studies, and the focus of this report by London Economics International LLC ("LEI"), is an assessment of the relationship between the actual locational marginal prices (LMPs) for energy in the PJM market and the short-run marginal cost ("SRMC") of producing electricity. The main task for this study was to estimate the effective prices (we refer to them as Perfect competition requires that the modeled LMPs through this report) assuming generators are following five parameters be fulfilled. offering their output exactly at SRMCs. Once short-run marginal · firms and consumers are price cost-based prices are estimated, they can then be compared takers; against actual historical LMPs and can be used to calculate a · the commodity is homogenous, i.e., price-cost markup index. there is no product


The study involves a two-stage process: (i) a simulation-based estimate of prices that would result if all generators in PJM Classic were bidding their SRMC and (ii) a comparison of those simulated (modeled) LMPs to actual market clearing prices in the Day-Ahead energy market. The ultimate analytical objective of this study is to estimate the price-cost markup index1 for the PJM Classic market area for the 43-month period, January 2003 through July 2006. The "cost" element of the price-cost markup index is specifically the shortrun marginal cost of the price-setting generator, which in the aggregate includes all costs that are variable across output levels in the short run2, such as fuel costs, variable operations and maintenance expenses, and emissions allowance purchase costs. The definition of SRMC used in this study focuses on physical operating costs, whereas some practitioners may also include opportunity costs. Indeed, PJM, in its implicit definition of reasonable marginal costs for purposes of bid mitigation also includes an adder for the recovery of other going forward costs.


there is perfect and complete information, all firms and consumers know the prices set by all firms; resources (including information) are perfectly mobile, so all firms have equal access to production technologies and capital; and there are no barriers to entry, any firm may enter or exit the market as it wishes.



If all the above conditions are present, then in such a market, prices would instantaneously move to an equilibrium, where firms make zero economic profits (only covering their costs, including their fixed and opportunity cost) and social welfare is optimized. In reality, however, many of the above conditions do not apply and therefore most markets depart to some degree from the theoretical premise of perfect competition.

The empirical analysis of this study can be characterized as comparing actual market dynamics to a theoretical benchmark based on the neo-classical economic theory of perfect competition. The basic tenets of economic theory predict that prices must equal SRMC under perfect competition (albeit in a hypothetical environment, because of the requirements


Price-cost markup index is defined as the ratio of the difference between actual LMP and estimated SRMC over the actual LMP. See Section 3.6 for further details and discussion. In this context, "short run" is defined as the period of time over which the plant's capacity and capital stock is fixed (i.e., prior to entire major maintenance changes or capital investments that could change the operating efficiency of the plant). -3London Economics International LLC 717 Atlantic Avenue, Suite 1A Boston, MA 02111


enumerated in the textbox to the right). Indeed, the complexity of the electricity industry with its barriers to entry and high fixed capital costs requirements may result in a situation that is not wholly congruent with the theory of perfect competition. Therefore, in the real world, the fact that prices will need to be above the short-run marginal costs so that firms can recover the minimum necessary going forward costs is a well accepted paradigm for energy-only markets. (Although it is important to note that PJM is not an energy-only market and operates other product markets, from which some generators can earn additional revenues.) From a policy perspective, the important question to ask is whether the observed markups above SRMC reported in this empirical study are adequate, too high, or too low to sustain the market. Although such policy applications are beyond the scope of the study, the results of this study provide important, statistically significant results that can be applied to think through and address such questions. Section 2 of the report provides a general overview of the PJM market place, and primarily describes the "Energy Market". Section 2 also highlights the details of the other product markets that PJM administers. A basic understanding of these other markets, namely, "Capacity" and the "Ancillary Services," is relevant to the study, because these other markets are a potential source of additional revenues for generators and therefore are important to understand from the prospect of analyzing the levels of the markups against fixed costs, and addressing questions of market efficiency, generator profitability, investment, and sustainability. Section 4 reviews the day-ahead locational marginal prices in their historical perspective within the modeled regions. Actual regional LMP is a basic component of the price-cost markup index; therefore it is important to understand the historical and geographical trends of these price series. Scope, methodology, and data As part of the scope of work, LEI proposed an analysis spanning a three and a half year timeframe starting from January 2003 through July 2006, coupled with a static geographical coverage area of PJM Classic (the original PJM service area). The selected timeframe is long enough to allow us to identify and study established trends in target variables. At the same time, the use of a static geographic designation permitted the results to be comparable across the timeframe and subject to direct contrast. In order to make the study computationally tractable, we selected a sample of days within each calendar year, representing the variations in load, seasons, and peak versus off-peak periods. Our sample size represents approximately 55% of the days in our timeframe leaving very small room for information loss that can otherwise occur with a small or biased sample. Given a sampled day, all 24 hours are simulated so that we get a full picture of daily variation of demand and supply of electricity across the day. In order to simulate SRMC-based prices, we needed first to estimate SRMCs, and then to simulate the price-setting process in PJM under the assumption of SRMC-based bidding by all generators. Section 3 describes the basic assumptions and methods used, including a description of the simulation model. Section 5 of this report explains how we estimated the short run marginal costs of the generating plants. In order to build the model, we gathered historical data that described market conditions at each hourly interval over the study timeframe. For example, we compiled actual load data for the selected PJM zones, as well as imports to and exports from the zones in question, transmission flows and limits through major transmission interfaces, fuel prices and other operating short run costs of the generating units, as well as detailed production data for major plants located in PJM Classic. We relied

-4London Economics International LLC 717 Atlantic Avenue, Suite 1A Boston, MA 02111

on data in the public domain as well as general accepted, industry sources to develop the SRMC-based bids and other necessary assumptions for the simulation modeling, such as the technical data on generating plants, published in FERC Form 1, FERC 423, EIA 906, IEA 423, U.S. EPA CEMS, U.S. EPA Clean Air Markets unit characteristics databases, collated and provided by Velocity Suite of Global Energy Decisions, actual zonal load data, hourly transmission limits, interchanges (imports/exports) from PJM, and historical spot market fuel price data from Bloomberg. The transmission constraints data (thermal limits and actual historical flows between transmission zones), which is crucial to determine the network topology, is not readily available publicly and was not provided by PJM. PJM, in contrast to other ISOs, also does not decompose its LMPs into an energy and congestion component (instead, PJM provides shadow-prices of constrained elements). Therefore, we had to rely on data on actual flows and limits on the major interfaces to construct our model's topology assumptions. The availability of data and PJM's own practices to monitor only the major interfaces led us to model PJM Classic as a five-region network. Summary of results Our results, which are discussed in Section 6, starting at page 43, show that for most of the months in the studied timeframe the price-cost markup indices, especially for peak periods, are significantly higher than zero3, indicating that actual average LMPs were higher than he modeled LMPs (which are based on estimated SRMC bidding). This is not surprising given the realities of market dynamics. PJM has acknowledged that it also believes that there is a positive markup above SRMC embedded in actual LMPs. However, the results of this study are interesting for the more subtle observations that are not apparent in PJM's markup indices. For example, the index levels vary significantly based on location (region), and time of day (peak versus off-peak) as well as across time. As expected, the peak period markup indices were almost always higher than the off-peak period markup indices for each region. This is intuitive given the differences in supply-demand balance during peak versus off-peak periods, and the types of resources that are price-setting as we move from off-peak demand levels to peak demand levels (and peaking resources' need for above-marginal cost bidding to recover going forward costs). Monthly markup indices for each region are quite volatile ­ standard deviations of indices for each region and year are close to half of the average indices for all periods and almost the same magnitude as the actual indices for the off-peak periods.4 For example, the average monthly markup index in the Delmarva peninsula area (region D) for peak periods is 10% with a standard deviation of 5%, and for off-peak periods the average is 6%, again with a 5% standard deviation over the study timeframe. In another example, the average monthly markup index in the service area of Pennsylvania Electric Company (region P) is 14% and 5% for peak and off-peak periods, respectively, both with a standard deviation of 4%. This volatility has a number of important implications. First, there is no apparent trend which correlates markup index levels with seasons or months. Therefore, other variables need to

3 4

In statistical jargon, most indices are statistically significant at the chosen 95% confidence level. In statistics, volatility of a variable is sometimes measured by the ratio of standard deviation and mean. If that ratio is higher than one third, the variable is considered "volatile" in statistical conventions, although volatility is an ordinal concept, used to compare variables rather than labeling them. -5London Economics International LLC 717 Atlantic Avenue, Suite 1A Boston, MA 02111

be explored to understand or explain the timing and levels of markup indices. Leaving aside potential structural changes in the market, since the magnitudes of standard deviations can be so close to the averages, any estimation of the markup index value outside of the studied timeframe (i.e. extrapolation to the future) faces a high degree of uncertainty. Therefore, we cannot conclude whether such markups are going to be repeated in the future from this historical backcast. Figure 1, below, portrays the range of monthly average5 price-cost markup indices for different periods across each of the years in our study timeframe (based on the sample days) and across modeled regions in PJM Classic.6 The difference between the market price and the short run marginal cost of the generating unit is called the price-cost markup, and the markup divided by the market price is called the price-cost markup index. Price-cost markup is measured in dollar terms, so it is difficult to compare two markup values, if the underlying prices are different. The price-cost markup index, in contrast, is free of units, and so allows temporal or across markets comparisons even though the underlying prices are different.7 We have focused on the markup index in this report, but have also augmented the discussions in Section 6 with the dollar-denominated markup levels (or differences) between actual LMPs and modeled LMPs, based on SRMC-bidding. The figure below shows the variation between peak and off-peak periods as well as a snapshot of the differences across regions within the same year. As discussed above and elaborated further in Section 6 of this report, these types of trends are (at least partially) explained by the supply and demand conditions within each region and across time. However, further analysis is required in order to robustly and exhaustively document the causes of explanatory variables of observed markups. The average on-peak monthly indices by region rarely exceed 20% over the study timeframe, with the range more typically below 10%, as seen below. Taking into account the forecast error of the model, some months' results become statistically insignificant at a 95% confidence level, and those are demarcated in the figure below with an asterisk.8


Throughout the report, unless it is explicitly stated otherwise, each period (hour, day, etc.) is weighed equally in calculating monthly averages, consistent with the Latin hypercube sampling method. See Section 3.4 for a detailed discussion of modeling PJM Classic topology. In short, PJM Classic has ten transmission zones based on the service areas of the local electric distribution companies. Given the transmission constraints in PJM Classic, we grouped them in five regions as follows: Region P: Pennsylvania Electric Company ("PENELEC"); Region M: 69% of the load of Metropolitan Edison Company ("METED") and PPL Electric Utilities Corporation ("PPL"); Region B: the remaining part of METED, Baltimore Gas and Electric Company ("BG&E") and Potomac Electric Power Company ("PEPCO"); Region E: American Electric Power Co., Inc. ("AECO"), Public Service Electric and Gas Company ("PSEG"), Jersey Central Power and Light Company ("JCPL") and PECO Energy Company ("PECO"); and Region D: Delmarva Power and Light Company ("DPL"). For example, consider two hypothetical electricity markets, one with a $100/MWh market price and a $90/MWh SRMC; and the other with $50/MWh price and $40/MWh SRMC. The price-cost markups in both markets are $10/MWh but the price-cost markup indices would be 10% and 20%, respectively (L.C. Section 3.6 at page 27 for the formula and discussion on price-cost markup index.). Note that Figure 1 only shows the range of markup indices with minimum and maximum values; detailed discussion of the statistical significance of monthly index values can be found in Section 6. -6London Economics International LLC 717 Atlantic Avenue, Suite 1A Boston, MA 02111




Figure 1. Annual ranges of the monthly averages of the price-cost markup indices for peak, off-peak and all periods across the modeled regions in PJM Classic

Peak Region P 2003 2004 2005 8% to 16% 9% to 20% 8% to 25% Off-Peak 2%* to 8% 1%* to 16% All 7% to 12% 6% to 15% 7% to 13% 1%* to 8% 1%* to 13% 4% to 7% 3%* to 7%* 3% to 10% 6% to 13% 5% to 11% 2%* to 6% 5%* to 16% 7% to 13% 3% to 10% 0%* to 8% 1%* to 15% 5%* to 19% 4%* to 11%

-3%* to 14% 3%* to 19% 0%* to 5%* 0%* to 9% 0%* to 5%* 0%* to 8% 1%* to 11% -1%* to 11% 1%* to 6% -1%* to 5%* 3%* to 14% 3%* to 14% 1%* to 4%* -1%* to 4%* 0%* to 13% 2%* to 14% 0%* to 5%*

2006 13% to 23% -1%* to 4%* Region M 2003 2%* to 13% 2004 2%* to 15% 2005 2006 6% to 15% 7% to 12% 5% to 12% 7% to 17% 7% to 16% 7% to 17% 8% to 19% 5% to 15%

-1%* to 12% 2%* to 14%

Region B Region E Region D

2003 4%* to 10% 2004 2005 2006

2003 2%* to 14% 2004 2005 2006

2003 1%* to 15% 2004 1%* to 17% 2005 2006 8% to 26% 8% to 16%

Note: 2006 is a partial calendar year (January ­ July 2006, only)

At the same time, due to the forecast error, which is region and time-specific, the average values that are computed in the study (and reported below) may in fact be higher (or lower) because at a 95% confidence level; there will be a higher bound (as well as a lower bound). In other words, a 11% markup index value for peak periods in January 2003 for region M with a 6% forecast error, corresponds to a confidence interval between 5% and 17%, hence we conclude the index in January 2003 for region M would take a value between 5% and 17% with a 95% probability (confidence level). The index value for the same month and region for off-peak periods is 2.5% and the forecast error is 5.5%, giving a confidence interval between -3% and 8%. Since the resulting confidence interval straddles 0%, we must conclude that off-peak index is not significantly different than zero (or statistically insignificant), and we cannot claim that there was a positive markup in that month for offpeak periods or average. Consideration of potential modeling critiques Modeling historical events in a region as large as PJM is a difficult task in and of itself, but is further complicated by the confidentiality or unavailability of primary data on transmission, technical operating constraints and generation plant characteristics over time. Because of the unavailability of detailed data on the transmission system (namely thermal transfer limits between the ten transmission

-7London Economics International LLC 717 Atlantic Avenue, Suite 1A Boston, MA 02111

zones in PJM Classic and congestion cost decomposition of LMPs), we simplified the topology and modeled a five region system within PJM Classic (plus all the interconnections with external markets and regions outside PJM Classic). This is an abstraction of the real world, as we are not modeling directly the internal congestion within each region or the entire nodal system. Nevertheless, we are calibrating generation and inter-regional flows to actual levels in order to offset the effects of this simplifying assumption. The magnitude of any possible bias is further minimized within each of our regions since we determine the boundaries of the regions carefully using the physical layout of the major transmission interfaces vis-à-vis generation and historical zonal price analysis (i.e., the zones with closer prices are grouped together in the same region). In summary, with careful preparation of regional designation points and calibration of modeling results, we believe that the simplifying assumption on transmission topology had limited effect on modeling results and the loss of information was minimized. A second simplification in the modeling relates to the generating unit characteristics. The key parameters were typically estimated hourly (for example, availability) or daily (spot fuel prices). However, we have also used monthly data (for hydroelectric production) or in some cases annual averages (for example, heat rates of individual plants) in our model. For plants where hourly generation data is unavailable, we have used generic technology based outage rates from North American Reliability Corporation's Generating Availability Data System (GADS)9 to project availability across the year. Though use of averages and generic industry data is not optimal, the operating parameters and technical characteristics for which we used monthly or annual data inputs do not fluctuate much typically and are not major drivers of the results. One possible critique of the modeling is that we did not model every day in the selected timeframe but only the selected days. The process of sampling almost always leads to some information leakage. In order to minimize problems, we used a very large sample size of 55%, and a robust statistical sampling technique to ensure that the sampled days represent the entire timeframe as closely as possible. In addition, we reviewed the days which were not sampled to understand if they had exhibited patterns of bias because they were outliers we did not cover. Based on an ex post examination of the load and LMP data, we have not excluded interesting periods, such as the hours with the lowest or highest actual LMPs. One other issue is the technical operating constraints faced by PJM controllers and system operators, which we did not take into account due to data unavailability. Many dynamic operating constraints, however, usually produce localized results and do not affect the average LMPs across the day or the regional aggregates. We did not account for certain plant-level operating parameters (for example, we did not consider ramp rates explicitly, although our proprietary power market simulation program, POOLMod,10 does have an ability to optimize dispatch through low-demand periods and so forth). In addition, partial outages and certain fixed costs (such as start up costs and no load heat costs) were not incorporated in the modeling, as we felt that they were already represented in the calibration process or were outside the scope of our definition of SRMC. Lastly, we also did not take into account actual

9 10

See for more information on GADS. See Section 8.3 for a detailed description of POOLMod. -8London Economics International LLC 717 Atlantic Avenue, Suite 1A Boston, MA 02111

hourly hydro production (because data at such a granular level is not available) but rather let POOLMod schedule the hydro subject to a daily energy budget and profit maximizing rules. There is also a methodological difference between our simulation modeling (and selection of the pricesetting unit) and PJM's mechanism in determining the LMPs. PJM's Security Constrained Economic Dispatch ("SCED") procedures use shadow prices ­ effectively the lowest undispatched bid price ­ in setting actual LMPs, while our proprietary simulation model looks at the highest dispatched price to set the modeled LMP. Theoretically, the difference between those two can be quite high, if the supply stack has big gaps. However, in a market as large as PJM Classic, the gaps in the supply stack tend to be smaller and thus negligible. Figure 59 through Figure 62 in Section 8.6 present the cumulative supply curves and demonstrate the fact that the gaps are indeed small. How to interpret the results? Every model, by definition, is an abstraction of complex, real systems; and therefore the results from any modeling require careful consideration, interpretation, and application to real world problems and policymaking. The fact that prices may depart from SRMC is not itself unusual or an indication of market dysfunction. Rather, it is a signal that further study and analysis is necessary before conclusions can be drawn about the efficiency of the market system in PJM. A closer look at the extent to which LMPs exceed marginal costs and an analysis of other income streams' contribution to fixed cost recovery is warranted to better understand how the markups relate to price levels necessary to motivate investment, and how the overall price levels relate to the long run marginal cost of the sector, or the break-even cost that is necessary to motivate investment and security of supply. Do markups exceed commercially reasonable profit levels for the sector and are they sustainable over a significant period of time, without erosion by new entrants? Do these price-cost markups indicate patterns in bidding behavior that demonstrate the potential for market power? This study does not attempt to answer all these complex questions, but rather provides the reader with the empirical basis for further examination.

-9London Economics International LLC 717 Atlantic Avenue, Suite 1A Boston, MA 02111


Microsoft Word - executive summary.doc

9 pages

Report File (DMCA)

Our content is added by our users. We aim to remove reported files within 1 working day. Please use this link to notify us:

Report this file as copyright or inappropriate


You might also be interested in

Microsoft Word - NY2-#657161-v2-Client_Alert_--_Good_Bank_Bad_Bank.DOC
Arctic National Wildlife Refuge (ANWR): Controversies for the 109th Congress
Wage Trends, 1800-1900
Interest Theory, Economics, and Finance