Beginning around June 2005, prices in the PJM day-ahead locational market pricing (DA LMP) energy markets and real-time pricing (RT LMP) markets rose precipitously. During most of 2004 and the first five months of 2005, the PJM monthly average unweighted energy prices generally did not stray too far from $40/MWh. However, prices finished 2005 in the $60/MWh to $80/MWh range. Nevertheless, in its recently released 450-page State of the Market Report for 2005, the PJM Market Monitoring Unit (MMU) found that “Energy market results were competitive.”1 The MMU found that fuel-price increases almost completely explain the increase in LMP.
What explains the PJM 2004/ 2005 price patterns and the run-up in the latter part of 2005? Was the run-up caused by market forces consistent with a well functioning competitive process, or is there evidence that market power inappropriately influenced market results? Does the PJM energy market, working in conjunction with other PJM markets, result in prices that are just and reasonable to both consumers and producers?
Based on publicly available information, our study concludes that price increases in PJM energy markets are not fully explained by higher loads and higher commodity fuel prices. Something else appears to be going on. Could higher energy prices be the result of the inappropriate exercise of market power rather than the appropriate result of market dynamics operating in the presence of scarcity?
Our basic finding is that, beginning in July 2005—one month after the price run-up began—actual DA LMP begins to exceed systematically its estimated value. Interestingly, load-serving entities in several state jurisdictions were conducting default service auctions during this period, including Delmarva Power on behalf of its approximately 20,000 Virginia ratepayers.2
Actual PJM energy prices observed in the latter half of 2005 no doubt influenced suppliers’ offers to sell electricity in those auctions. Since the results of these auctions directly translate into much higher retail prices for electricity, it is important to determine why actual prices exceed estimates in what appears to be a systematic way during the last six months of 2005.
But there is a crucial flaw in the analysis. While PJM has provided the fuel type of the marginal unit, the information that is needed to increase the precision of the analysis and corresponding faith in any set of results is the heat rate of the marginal unit(s). This data is not yet publicly available.3 Nevertheless, by keeping heat-rate assignment rules constant throughout the 24-month period, it is posited that study of the available data results in increased understanding of PJM energy market results during this especially important time period.
The results are very interesting. For many sets of heat-rate assignments tested, actual DA LMP begins to rise relative to estimated DA LMP beginning in July 2005. This pattern holds through the end of the study period. December 2005 has by far the highest actual DA LMPs relative to estimated DA LMPs for any month in the study period. Again, since some states have been conducting auctions during the last half of 2005, the above result is very troubling for consumers if it is an indicator of the inappropriate exercise of market power.
Before proceeding to analytical details, a short discussion of institutional constraints is in order. Some may object to this attempt to push forward with analysis that is admittedly institutionally challenged. In response, note that this is the only recourse available to independent analysts of organized markets. Market advocates claim that, to protect the ability of markets to function, PJM and other RTOs must closely guard key data and information. This necessitates a tradeoff between the often conflicting goals of allowing markets to work on the one hand and instilling confidence that they are working correctly on the other.
Figure 1 shows the problem. Beginning in June 2005, PJM DA LMP begins a precipitous rise to almost $80/MWh from levels about half that for much of 2004. Corresponding prices for January and February 2006 are $50.54/MWh and $53.42/MWh, respectively. Real-time locational marginal prices exhibit a similar pattern. Figure 2 adds estimated DA LMP values as calculated in this study. As noted, one month after actual DA LMP prices began their climb, actual DA LMP begins to systematically exceed its estimate. These prices are arithmetic averages across all hours; load weighting would increase all reported values.
At this point, the reaction to this price run-up pretty much can be predicted by who is doing the reacting. Competition advocates, generators, and some regulators who have supported industry restructuring designate hot weather and fuel prices as the primary causes of the observed prices. On the other hand, load interests, consumer advocates, and regulators who have been critical of industry restructuring complain that markets simply do not and cannot work, and that extraordinary relief must be forthcoming.
Is one camp more right than the other? Is there any information and data available for independent analysis that can shed any light on this important question?
In the fall of 2005, at the request of the Virginia State Corp. Commission staff, PJM posted marginal fuel-type data on its Web site for each hour of 2004. Monthly files contain data on the fuel types of the marginal (i.e., the last and most expensive offered) generation unit selected to meet load in any particular hour. Each hour is divided into 12 five-minute intervals. When there is no congestion on the PJM system for an entire hour, a single fuel-type is marginal for the entire hour. If there is no congestion but the marginal unit changes during the hour and the latter unit is fueled from a different fuel-type, two different values are reported for the hour based on how many five-minute intervals each fuel was on the margin. When there is congestion, there can be more than one marginal unit during a particular five-minute interval. As PJM describes on its Web site, the share of each fuel in each hour is calculated based on the number of five-minute intervals that a unit burning each fuel type is marginal or jointly marginal.
Although data for 2004 had been available for several months, questions arising from the 2005 LMP price run-up inspired Virginia commission staff to request that data for 2005 be posted to the Web site or otherwise provided. Quickly responding to that request, PJM posted year-2005 marginal fuel-type data on Jan. 31, 2006. The data contains information indicating when coal, natural gas, light oil, heavy oil, landfill gas (LFG), wind, and interface power are running on the margin. “Miscellaneous” also is reported as a fuel-type category. The vast majority of hours had some combination of coal, natural gas, or petroleum as the marginal unit. LFG, wind, and interface appeared extremely infrequently and were dropped from the analysis.
Because heat rates are not reported, “miscellaneous” also was dropped. Heavy and light oil were combined into a single category. Table 1 summarizes the frequency of coal, natural gas, and oil as the marginal fuel. Changes in the PJM footprint significantly affected the percentage that each fuel was marginal during the two-year study period. Commonwealth Edison (Com Ed), AEP, Duquesne Light (DLCO), and Dominion Virginia Power (DVP) integrated their generation and transmission systems on May 1, 2004, Oct. 1, 2004, Jan. 1, 2005, and May 1, 2005, respectively.
The first analytical task is to merge spot-fuel price data with marginal fuel-type information so that hourly LMP may be estimated. The calculation is based on which fuel is running on the margin, the portion of the hour that the fuel was marginal, spot-fuel prices,4 and heat-rate assumptions. Monthly results as plotted in Figure 2 are set forth in Table 2.5 The hourly estimation procedure calculates weighted average LMP by adding across each marginal fuel type’s contribution to that hour’s total. Each fuel type’s component is calculated as appears below:
In Table 2, only the fuel type of the marginal unit is known; the heat rate of that marginal unit remains a mystery.6 This problem most seriously manifests itself in the assignment of the proper heat rate when natural gas is indicated as the marginal fuel. Is a combined-cycle combustion turbine (CT) with a relatively low heat rate dispatched as the marginal unit or is a relatively inefficient simple-cycle CT setting the price? The analysis crucially depends on which gas technology is assumed to be marginal when natural gas is indicated as the marginal fuel.
First, we asked PJM to provide the heat rates of the marginal units and waited for a response. Next, we pressed ahead by assigning heat rates as best we could. For example, when we encountered a relatively high load hour (relative to that month’s peak load) during winter or summer, a simple-cycle CT (12,000 Btu/kWh) was assumed to be on the margin when natural gas is indicated as the marginal fuel. If it is the wee hours of the morning, a shoulder month, or a relatively light load hour, a combined-cycle CT (7,500 Btu/kWh) was assumed to be running on the margin. The important feature of the study is that the heat-rate assignment rules are held constant during the 24-month study period. This means that the pattern of the results is interesting, not necessarily the excess or deficit in any particular month.
Given these caveats, the results as shown in Figure 2 indicate that beginning in July (not June, curiously) 2005 and lasting through November, actual DA LMP began to exceed its estimate by about $15/MWh more than had been evident during the preceding 14-month period. Prices in December exceeded their estimate by an incremental $27/MWh versus the first 18 months of the study. Note the words “more” and “incremental,” which make clear that, due to this analysis’ admitted deficiencies, we cannot be certain of the absolute level of the difference between DA LMP and its estimate. Rather, the interesting finding of the study is the change in the pattern beginning in July 2005.
What is the import of these results? First, if an additional $27/MWh were generated in the PJM energy market in December 2005 and found its way into the hands of generators, this would provide an additional $1.7 billion in just that one month alone. The $15/MWh overage in July through November represents another $4.7 billion. This, plus the December overage of $1.7 billion, yields an “extra” $6.4 billion to generators during the last half of 2005. This works out to about $40/kW for each of PJM’s approximately 160,000 MW of generation capacity.
Figure 3 contains the daily difference between actual DA LMP and estimated LMP over the two-year period. The difference is defined as actual DA LMP minus estimated DA LMP.
The daily results again show that, beginning in July 2005, actual DA LMP begins to exceed estimated LMP in a rather extraordinary way, at least as compared with the first 18 months of the time period.
We constructed an econometric model that attempts to explain the variability in this difference over the study period. As might be expected, preliminary results show that the difference between actual DA LMP and estimated DA LMP is positively related to the level of DA LMP itself. An interesting future project would be to construct a more sophisticated model to see how this difference (actual LMP minus estimated LMP) through time relates to fuel prices, season of year, time of day, PJM geographic configuration, marginal fuel, loads, generator outages, or other variables.7
What to make of all of this? The PJM MMU’s State of the Market Report for 2004 found that most PJM energy, capacity, and ancillary services markets were competitive in 2004.8 The MMU makes a similar finding for 2005, specifically finding that energy market results were competitive. The 2004 report notes:
“Market structure issues in the PJM Energy Markets have been offset to date by a combination of high levels of supply, moderate demand and competitive participant behavior.”9
Market structure generally refers to measures of generation ownership concentration quantified by calculations of Herfindahl-Hirschman Indices (HHI). The 2004 report notes some concerns about high levels of market concentration in general as well as in particular geographic areas. Apparently, the MMU was concerned with the high degree of market concentration in PJM energy markets, but noted that because of high levels of supply, moderate demand in 2004 and “competitive participant behavior” prices remained competitive. The 2004 report goes on to state:
“No evidence exists, however, that market power was exercised in these areas during 2004, both because of generator obligations to serve load and because of PJM’s rules limiting the exercise of local market power.”10
Interestingly, after stating similar concerns, the 2005 report identically states:
“No evidence exists, however, that market power was exercised in these areas during 2005, both because of generator obligations to serve load and because of PJM’s rules limiting the exercise of local market power.”11
The above comment about market power being constrained by “generator obligations to serve load” is crucial if it is indeed correct. It may help explain the results of the study set forth in this paper. Since about 80 percent of the energy generated in PJM in 2004 was produced by entities that must also supply customers at prices that are constrained by some form of regulation, the financial benefit to generators that would otherwise accrue from high energy prices might have been limited in the past. This would have depended on specific circumstances in particular jurisdictions. Nevertheless, generator obligations to serve load at capped rates was thought to have had a calming effect on prices in the past.
Such calming influence may wane as the obligation to serve retail load at regulated prices and the generation needed to serve that load are financially divorced. Generation that no longer must, in some way, share margins with regulated customers would be able to unilaterally reap the benefits of high market prices. In the many jurisdictions where auction results directly translate into increased prices for customers—and profits for generators—the divorce is fast becoming effective. Although the 2005 report finds to the contrary, could near-term diminution in generators’ obligation to serve load at capped prices be a factor in the 2005 PJM price run-up?
State regulatory commissions must try to participate in these matters as best they can. This is even more true in states where rate caps are coming off and retail customers face the “sticker shock” of market rates even as they recall the promises that changes in this industry were supposed to reduce electric bills.
It also is troubling that the result obtained by this analysis shows that energy market margins expanded greatly in the latter half of 2005—at odds with the PJM MMU’s finding that the price-cost markup index did not vary greatly over 2005.12 In fact, the MMU found that relatively small margins tended to get smaller as year 2005 progressed. It would be nice to be able to reconcile these two results.
We suggest that net cash flows to all existing generators have increased in a large and systematic way beginning in June of 2005. Corresponding price increases are thought to affect supplier offers of electricity (both price and quantity) into default service auctions and, as such, are effectively passed on to retail customers. It is far from clear the degree to which market power is exerted in default service auctions themselves or the PJM energy markets on which the default service auctions are based. Whether or not market power is being exerted inappropriately, generators’ increased cash flows should not be ignored by FERC, other policy-makers, and stakeholders. F
1. 2005 PJM State of the Market Report at p. 23; released March 8, 2006.
2. For example:
a) Delaware: For customers of Delmarva Power; RFP issued in August 2005 with bids opened in January 2006 resulting in a reported 59 percent increase for residential customers on May 1, 2006.
b) New Jersey conducted an auction in early February 2006 for one-third of its retail default service obligation. The results of the auction are expected to result in about a 13 percent increase for retail customers.
c) The District of Columbia is about to conduct an auction, results of which will be reflected in retail rates beginning on June 1, 2006.
d) Maryland issued an RFP in December 2005 with bids due in February 2006. Retail prices, if not mitigated, are expected to result in retail rate increases of between 50 percent and 80 percent depending on service class and distribution utility.
e) Pike County (PA) Light & Power raised rates approximately 70 percent on Jan. 1, 2006, as a result of an auction process that began in May 2005 and concluded in October 2005. Note: As Pike County L&P is an Orange and Rockland (NY) affiliate, its service territory is in the NY-ISO.
3. Just as this paper was being finalized, PJM responded to an informal request for the heat rate of the marginal unit by stating that it does not collect such data. As such, it is not available according to PJM’s initial response. As misunderstandings and miscommunications are always possible, this data request will be pursued.
4. Fuel prices used were Intercontinental Exchange Transco Zone 6 (NY) for natural gas, NYMEX spot coal, and Intercontinental Exchange No. 2 fuel oil.
5. For hours where oil, coal, and natural gas did not represent 100 percent of the hour, the estimated LMP for the hour was “grossed-up.” For example, if oil, coal, and natural gas represented 90 percent of a particular hour’s marginal fuel, the result calculated for 90 percent of that hour was multiplied by 1/.9 or 1.11. Since estimated LMP is calculated as a weighted-average, the gross-up procedure ensures that hourly weights sum to unity.
It also should be noted that July through October 2005 had a relatively large incidence of “miscellaneous” appearing on the data set. Further investigation revealed that “miscellaneous” tended to occur during relatively lower-load hours. As such, the gross-up procedure is thought to be conservative in those months since it is thought that the underlying marginal unit would tend to have lower marginal cost than the cost being grossed up. November and December 2005 had relatively few “miscellaneous” entries.
6. As noted in the text, the biggest single deficiency in this analysis is that the heat rates of marginal units, while held constant through the 24-month study, are not known with certainty. Another deficiency in this work is that emissions costs are ignored, but they are ignored for the entire two-year period.
7. DA LMP was chosen as the primary price indicator for this study. It was thought that, although highly correlated, DA LMP would have significantly less variability than RT LMP. Nevertheless, regression modeling included a variable calculated as the difference between these two price measures. Early results indicate that the difference between DA and RT LMP do not greatly explain the variability in the difference between DA LMP and its calculated estimate.
8. 2004 PJM State of the Market Report at p. 20.
9. 2004 PJM State of the Market Report at p. 22.
10. 2004 PJM State of the Market Report at p. 46.
11. 2005 PJM State of the Market Report at p. 23.
12. The price-cost markup index is defined by the PJM MMU as the difference between price (P) and marginal cost (MC), divided by price, where price and marginal cost are determined by the offers of the marginal unit [The markup index = (P – MC)/P]. The marginal unit is the unit that sets LMP in the five-minute interval. The markup of each marginal unit is load-weighted. The markup index is normalized and can vary from -1.00, when the offer price is less than marginal cost, to 1.00, when the offer price is higher than marginal cost. See 2005 PJM State of the Market Report at p. 83.