Pricing Logic Under Flexible Modeling of Constrained Output Generating Units

Draft Final Proposal Pricing Logic Under Flexible Modeling of Constrained Output Generating Units April 14, 2008 This is the third CAISO paper on this issue. An Issue Paper was posted on February 1, 2008 and a Straw Proposal was posted on March 7, 2008. These are located at http://www.caiso.com/1822/1822931f287d0.html

Pricing Logic Under Flexible Modeling of Constrained Output Generation (COG) Units Table of Contents 1 Introduction... 3 2 Process and Timetable... 4 3 Key Criteria for Evaluating Potential Solutions... 5 4 Stakeholder Feedback... 5 5 Proposed Design Solution... 6 6 Conclusion... 8 7 References... 9 Appendix... 10 MPD/GVB April 4, 2008, page 2

Pricing Logic Under Flexible Modeling of COG Units Prepared for Discussion on the Stakeholder Conference Call on April 21, 2008 1 Introduction Constrained Output Generation (COG) units are those that are inflexible or lumpy in that they must operate at their full maximum output levels when they run. In addition, COG units generally have minimum run times such that, once they are started, they cannot be shut down until a pre-specified number of intervals has passed. While these units tend to be expensive to operate, they are often able to ramp up quickly and thus can fill an important gap in meeting peak demand, and can also quickly relieve shortages due to forced outages. In light of this, and following a FERC directive to do so, the MRTU market design includes the feature that COG units be able to set the LMPs. [October 28, 2003 FERC Order, [1]] To accomplish the objective of enabling COG units to set LMPs, the CAISO will model COG units as flexible resources under MRTU upon the start-up of the MRTU markets (i.e., Release 1 ). 1 This is expected to result in greater consistency between market energy prices and the operating needs of the CAISO system, and thus will provide good incentives for increased participation in the CAISO markets by the COG units as well as sending more accurate price signals to all market participants. Not all of the peaking units in the CAISO control area can opt to be considered COG units. To assess the magnitude of potential issues with modeling COG units as flexible in the RT Pricing Run, the CAISO has endeavored to determine the population of units that could elect to be treated as truly lumpy under MRTU. The CAISO used, as a starting point, all the gas turbine (GT) resources within the CAISO Master File. 2 Then, units that are not dispatchable by the CAISO were ruled out. Non-dispatchable units include those that are Qualifying Facilities, Must Run or Must Take, and some units that are part of Metered SubSystems. Additionally, GT units for which the difference between Pmax and Pmin was large were ruled out, as these units would not meet the definition of COG units under the MRTU Tariff. 3 After narrowing the population of potential COG units based on these criteria, there remain approximately 10 to 15 potential COG units with a total generating capacity of 250 to 350 MW in the CAISO master file. This aggregate generating capacity represents roughly 1% of average peak hour load during the summer of 2007. 4 The units are spread evenly throughout the control area. 1 2 3 4 In fact, COG units are modeled as flexible resources in the IFM in both the scheduling and pricing runs, but in Real Time (RT), they are modeled as flexible only in the pricing run. The RT scheduling run respects actual operating constraints (that is, lumpiness) in order to ensure a feasible RT dispatch. This is consistent with FERC guidance, practices at other ISOs, and input from CAISO Operations Staff. Section 27.7 of the current MRTU Tariff addresses the definition and treatment of COG units. This percentage is based on average Hour Ending 16 load for the CAISO control area for June through September 2007, which is 37,055 MW. MPD/GVB April 4, 2008, page 3

The issue being examined through this stakeholder process is the extent to which refinements to the COG pricing logic are needed and, if they are needed, then how those refinements should be designed. The issue paper on this topic outlines distortions that can result from the flexible modeling of constrained units. The straw proposal offered several options for changes to the market software that would mitigate the temporal or stuck price issue which was identified as the most concerning of the potential distortions. 5 Following stakeholder comments, MSC discussion, and further analysis, the CAISO recommends that no changes be made to the market software in Market Release 1A, and offers this draft final proposal to this effect. In large part this recommendation is driven by the small number of COG units, and the fact that additional refinements made to the pricing logic may well have more far-reaching deleterious pricing ramifications. 2 Process and Timetable The purpose of this proposal is to reflect stakeholder input and CAISO analysis on the issues described is the issue paper, as well as on the potential design change options described in the straw proposal. Questions and comments are welcome and encouraged. Based on any additional stakeholder and MSC input on this draft final proposal, the CAISO will develop a final proposal which will be submitted to the CAISO Board of Governors in July. Any changes to the MRTU market rules that are developed as a result of this process will be included in the Market Release 1A package to be launched no later than one year after the start of MRTU. The table below summarizes the key steps in the stakeholder process on COG pricing. Note that, at the request of some stakeholders, the CAISO is extending the time for review and resolution of this issue. CAISO staff is targeting the July 9-10 Board meeting for resolution of this issue, and approval of any design changes. February 1 Post Issue Paper February 8 MSC/Stakeholder meeting February 22 Stakeholder comments due * March 7 Post CAISO Straw Proposal March 17 Stakeholder conference call March 24 Stakeholder comments due * April 14 Post Draft Final CAISO Proposal April 21 Stakeholder conference call April 30 Stakeholder comments due * May 9 Post Final CAISO Proposal July 9-10, 2008 Presentation to CAISO Board of Governors 5 The temporal price issue was identified by LECG, [2]. MPD/GVB April 4, 2008, page 4

* Please e-mail comments to Gillian Biedler at gbiedler@caiso.com 3 Key Criteria for Evaluating Potential Solutions This section provides some key evaluation criteria the CAISO believes are important, and invites stakeholders to identify other criteria that should be considered in assessing potential solutions. Any policy that is developed should balance the objective of correcting the temporal and spatial pricing inconsistencies noted above with providing constrained units the needed incentives to participate fully in the CAISO markets and sending all market participants price signals that reflect the cost of utilizing COG units to meet the operating needs of the system. Any policy that is developed should balance the need for uplift to COG units (when treated as lumpy and ineligible to set price) with the uplift payments due to flexible resources that are dispatched down in RT to make room for COG units scheduled as flexible in the IFM. Policy and design options should be evaluated for implementation feasibility and costs for both the CAISO Stakeholder and for the CAISO. This evaluation should be done keeping in mind the magnitude of the potential issue, i.e., in light of the fact that there is a relatively small number of COG units in the control area. 4 Stakeholder Feedback In the straw proposal posted March 7, 2008, there were two main elements that the CAISO identified for consideration when evaluating potential market design changes to better deal with the issues presented by COG units. These potential changes were (1) to change the RT Pricing Run initialization for non-cog units, and/or (2) to refine Real Time pricing outcomes by changing the RT optimization or adding a post-market adjustment to prices. These potential design changes were discussed in detail in the straw proposal. The CAISO appreciates the feedback of stakeholders on this issue. Following, please find summaries, by participant, of the feedback received. Pacific Gas and Electric (PG&E) offered tentative support of the Status Quo design option. In order to fully endorse this option, PG&E requested some additional analysis. Specifically, they asked for (1) an assessment of the number of hours expected to be impacted by the stuck price problem, (2) examples demonstrating the benefits and drawbacks of the pricing run initialization change with and without uninstructed deviations, (3) review of this proposed design by LECG, and (4) monitoring of the stuck price issue after MRTU start-up. MPD/GVB April 4, 2008, page 5

The frequency with which a COG unit would have been marginal had it been eligible to set the price is not known since that functionality is not part of the current CAISO market design. In an effort to address the question as best as possible, however, the percentage of hours in which a sample of lumpy COG units operated in 2007 was analyzed. In particular, data for the units that have opted not to waive COG status under MRTU market simulation show that each of those units ran on average about 5% of hours during the entire year. Clearly, however, the periods in which these units would be marginal would likely be significantly less than the number of hours in which the unit ran during 2007. This suggests that the opportunity to set the price would have presented itself a maximum of 438 hours per unit over the course of the year. It is important to point out that the price a COG unit could set will be based on its Minimum Load bid divided by its Pmin rather than on a submitted energy bid. The CAISO agrees that examples, however simple, are very helpful in understanding these technical issues. The requested examples are included in the appendix to this Draft Straw Proposal. The CAISO will provide LECG with documentation of the proposed design for COG pricing under MRTU and share any relevant feedback with stakeholders. The CAISO commits to monitoring COG participation in the MRTU markets, with specific attention to those instances in which COG units set the Market Clearing Price. Southern California Edison (SCE) also offered support for the Status Quo design option with the recommendation that the CAISO monitor COG pricing and performance once MRTU data are available. The CAISO concurs and, as noted above, it committed to monitoring upon MRTU start-up. Western Power Trading Forum (WPTF) supports the Status Quo design option and also expressed support for monitoring of COG unit pricing under MRTU. Analysis of pricing outcomes under the other market design options described in the Straw Proposal may prove infeasible, but this suggestion is appreciated and will be kept in mind. 5 Proposed Design Solution The CAISO proposes that no changes with respect to COG pricing be made as part of Market Release 1A. The current MRTU design is that a unit that is strictly lumpy, and thus does not have a dispatchable range greater than 3 MW or 5% of its Pmax, will be modeled in the Day Ahead scheduling and pricing runs, and in the Real Time pricing run as fully flexible. In the Real Time scheduling run, the unit will be modeled according to its true operating constraints in order to ensure a feasible dispatch solution. In those market runs for which the COG is treated as fully flexible, a fictional dispatchable range from 0 to Pmax is imposed, which makes it more likely that the unit will be able to set the Market Clearing Price. A strictly lumpy COG unit MPD/GVB April 4, 2008, page 6

cannot, however, submit an energy bid. The price at which it offers energy into the market is calculated as its Minimum Load Bid divided by its minimum operating level (Pmin). This Minimum Load Bid is either a cost-based or bid-based value submitted to the Master File. While not dynamically mitigated, it is planned that Minimum Load Bids will be frozen for six months, and subject to certain caps. 6 This is anticipated to be sufficient price mitigation according to all analysis and feedback. In the straw proposal, two main design elements were proposed. These potential changes were (1) to change the RT Pricing Run initialization for non-cog units, and/or (2) to refine Real Time pricing outcomes by changing the RT optimization or adding a post-market adjustment to prices. The second design element was deemed to be very difficult if not impossible from an implementation perspective. Even if possible to implement, this would represent a major change at great effort and expense for a very small number of COG units. Stakeholder feedback and preliminary MSC discussion support the CAISO recommendation that none of the three possibilities (2b, c and d) put forth as potential design changes in the straw proposal be undertaken. The change to the RT Pricing Run initialization was recommended by LECG in its 2005 review of the MRTU market design, [2]. This recommendation cited the fact that the NYISO changed its pricing run initialization to enable the 4,800 MW of capacity provided by COG units to be more readily replaced as price-setters by less expensive units. While the pricing run change has had this positive effect in the NYISO, it has also led to considerable issues when non-cog units deviate from dispatch instructions. Specifically, Real Time prices are based on the pricing run which, given this change, assumes that units are responding to dispatch instructions, while the units themselves may be operating quite differently from those instructions for a variety of reasons. While the change to the pricing run initialization does not create the incentive to deviate from dispatch instructions, uninstructed deviations lead to wider disparities between operating conditions and prices when the pricing run is not tied to telemetry. In short, Real Time prices are not necessarily reflective of actual conditions when pricing is not tied to telemetry and this problem is exacerbated in the case of uninstructed deviations. The philosophy that Market Clearing Prices be determined at the time of the market run in order to send timely price signals that reflect contemporary operating conditions is fundamental to the MRTU design. To further illustrate the issue of inaccurate prices in circumstances involving uninstructed deviations, please refer to the examples included in the appendix to this paper. Note that examples can be constructed in which the incorrect prices are either higher or lower than accurate prices. The concern is not one of prices that are systematically too high or low, but of prices that are derived from projected operating conditions rather than being based on actual conditions. In both the NYISO market and in the CAISO under MRTU, this presents a difficulty because prices are not adjusted ex post as they are in PJM and ISO-NE, for examples. There is not the opportunity, therefore, to adjust prices to reflect these uninstructed deviations. Furthermore, ex post adjustment of prices is counter to the philosophy under which MRTU was designed. To 6 More information about the mitigation of Start Up and Minimum Load Bids is available at the following link to the September 6-7, 2007 CAISO Board of Governors meeting: http://www.caiso.com/1c49/1c49f43c28e00.html. This proposal was filed with FERC on October 19, 2007. At the writing of this proposal, a ruling had not yet been made on this filing. MPD/GVB April 4, 2008, page 7

reiterate, prices that do not reflect operating conditions because they are based on a projection from the previous pricing run rather than telemetry is counter to this principle underlying the MRTU design. Stringent uninstructed deviation penalties are in effect in the NYISO market, but are not fully effective at deterring these deviations due to myriad factors such as bilateral contractual arrangements. The MRTU market design does not currently have uninstructed deviation penalties in place, and so the potential exposure to this issue in California would be even greater than it is in New York. In the NYISO, operator intervention is required to help remedy situations in which uninstructed deviations are causing reliability issues. The drawbacks to implementing the change to the pricing run could be frequent since any one of a large population of generating units could deviate from its dispatch instructions. The benefits are likely to be far rarer as they would only be realized when a COG unit of which there are very few was not displaced as quickly as it could be by a less costly unit. This cost/benefit assessment, in addition to the argument based on the principles underlying the MRTU design, support the CAISO recommendation that the change to the pricing run initialization not be implemented. It is not anticipated that instances in which the stuck price problem occurs will be frequent or long lasting. It is impossible to know with any certainty, of course, until some experience with the MRTU market is gained. For this reason, the CAISO intends to monitor for instances in which a COG unit sets the LMP, and also for intervals in which it would be eligible for uplift as it is running but not setting the LMP. Though this won t tell us exactly how many intervals experienced the stuck price problem, it will give us an upper bound. The precise threshold at which the number of intervals exhibiting this problem becomes problematic is unclear, but certainly routine monitoring will help shed light on expected frequency and duration, and also any changing trends in the frequency or duration of the stuck price problem. 6 Conclusion The flexible modeling of COG units has the advantage of enabling them to set the LMP when they are needed to meet system conditions. This will help to provide those units with incentives to participate more fully in the CAISO markets. Based on analysis by CAISO staff, the preliminary input of the MSC, feedback from stakeholders, and considering the small number of potential COG units in California, the CAISO recommendation is that no changes be adopted in Market Release 1A relative to COG pricing. CAISO staff commits to performing routine monitoring under MRTU to evaluate the potential frequency and magnitude of any inter-temporal ( stuck ) price issues resulting from COG unit constraints to revisit the need for market design changes in response to adverse indications from this monitoring. The CAISO is seeking to balance the importance of accurate price signals with the implementation costs and difficulties associated with various design change options. It is very important when evaluating this trade-off to consider the fact that there are currently very few COG units with a small aggregate generating capacity in the California ISO control area. Stakeholder questions and comments are welcome and encouraged. MPD/GVB April 4, 2008, page 8

7 References [1] FERC Docket #: ER02-1656-004 issued 28 October 2003, Further Order on the California Comprehensive Market Redesign Proposal [2] Law & Economics Consulting Group (LECG), February 23, 2005, Comments on the California ISO MRTU LMP Market Design Section V. Real Time Dispatch, B. COG Pricing (pp 60-62) MPD/GVB April 4, 2008, page 9

Appendix The following examples are designed to illustrate the phenomenon that prices can fail to reflect operating conditions when the pricing run is not based on telemetry. The first and second examples show simplistically how the CAISO MRTU optimization would arrive at pricing and dispatch outcomes under the change to the pricing run initialization in the scenarios with and without uninstructed deviations. The third example demonstrates what prices and dispatch would be if the pricing run initialization were not changed and there were uninstructed deviations. A vital point is that this outcome is completely separate from whether or not a COG unit is needed. Changing the pricing run to be based on the previous pricing run outcome would impact all units and time periods and thus this disconnect between prices and operating conditions could occur whenever the telemetry and the pricing run outcomes do not match. In both situations with uninstructed deviations, telemetry indicates that there is over-generation. This is highlighted in orange below. The extent to which operators would intervene, generators might be backed down, or exports might be increased in order to deal with this situation is outside the scope of this admittedly simple example. Note, however, that this abstraction does not detract from the fundamental point of the examples which is that prices based on the previous pricing run will reflect the same information about system conditions as do prices based on telemetry. Also, please note that examples can be constructed in which prices are either lower or higher as a result of uninstructed deviations. In the examples presented below, the price in period t+2 is higher when prices are based on telemetry than they would be if based on the previous pricing run given positive uninstructed deviation by unit Flex2. Finally, the arrows in the tables below are included to indicate which information is being fed forward into the next period s pricing run. To reiterate, in the first table, the pricing run resolution coincides with telemetry. In the Second table, the pricing run solution no longer coincides with the telemetry which reflects the uninstructed deviations. The third table shows the pricing result when the telemetry data with the uninstructed deviations are carried forward. The scheduling run is always based on the previous period s telemetry data. MPD/GVB April 4, 2008, page 10

Givens: COG GT that operates at 0 or 50MW, and has a Pmin/ML Cost = $100 per MWt, with a minimum run time of two intervals Flexible unit 1 has range [0,255] and a bid price of $40/MWt Flexible unit 2 has range [0,100] and a bid price of $60/MWt Flexible unit 1 has a ramp rate of 10 MW per t, and Flexible unit 2 has a ramp rate of 35 MW per t The CAISO Approach Change to pricing run initialization, WITHOUT uninstructed deviation: Scheduling = Time Load Unit Telemetry = Dispatch Initialization based on Previous Pricing Run Pricing COG 0 0 t-1 235 Flex1 235 235 Flex2 0 0 COG 50 30 t 275 Flex1 225 235 +/- 10 => 225 to 245 245 Flex2 0 0 +/- 35 => 0 to 35 0 LMP $100 COG 50 0 t+1 270 Flex1 215 245 +/- 10 => 235 to 255 255 Flex2 5 0 +/- 35 => 0 to 35 15 LMP $60 COG 0 0 t+2 245 Flex1 225 255 +/- 10 => 245 to 255 245 Flex2 20 15 +/- 35 => 0 to 50 0 COG 0 0 t+3 245 Flex1 235 245 +/- 10 => 235 to 255 245 Flex2 10 0 +/- 35 => 0 to 35 0 COG 0 0 t+4 245 Flex1 245 245 +/- 10 => 235 to 255 245 Flex2 0 0 +/- 35 => 0 to 35 0 In t, the COG is needed to meet a load increase and thus sets the MCP. In t+1, the Flex2 unit is marginal and sets the LMP. The COG remains on due to its min run time constraint. By t+2, the Flex2 unit is displaced by the least-cost Flex1 in the pricing run. Flex1 is able to do so because its eligibility is based on the t+1 pricing run quantity rather than telemetry. Were this not the case, its ramping constraint would prohibit it from reaching a high enough output to displace Flex2. The CAISO Approach Change to pricing run initialization, WITH uninstructed deviation: Time Load Unit Telemetry Scheduling = Dispatch Initialization based on Previous Pricing Run Pricing t-1 235 Flex1 235 235 235 Flex2 0 0 0 COG 50 50 30 t 275 Flex1 225 225 235 +/- 10 => 225 to 245 245 Flex2 15 0 0 +/- 35 => 0 to 35 0 LMP $100 COG 50 50 0 t+1 270 Flex1 215 215 245 +/- 10 => 235 to 255 255 Flex2 30 5 0 +/- 35 => 0 to 35 15 LMP $60 t+2 245 Flex1 225 225 255 +/- 10 => 245 to 255 245 Flex2 45 20 15 +/- 35 => 0 to 50 0 t+3 245 Flex1 235 235 245 +/- 10 => 235 to 255 245 Flex2 35 10 0 +/- 35 => 0 to 35 0 t+4 245 Flex1 245 245 245 +/- 10 => 235 to 255 245 Flex2 75 0 0 +/- 35 => 0 to 35 0 In t, the COG is needed to meet a load increase and thus sets the MCP. In t+1, the Flex2 unit is marginal and sets the LMP. The COG remains on due to its min run time constraint. By t+2, the Flex2 unit is again displaced by the least-cost Flex1 in the pricing run. It does not respond to dispatch instructions. The MCP of $40 in this period does not reflect the fact that Flex2 is generating at a level that makes it physically infeasible for Flex1 to serve all of load. Flex2 is marginal based on telemetry, but not on pricing run output. The CAISO Approach NO change to pricing run initialization, WITH uninstructed deviation: Time Load Unit Telemetry Scheduling = Dispatch Initialization based on Telemetry Pricing t-1 235 Flex1 235 235 235 Flex2 0 0 0 COG 50 50 30 t 275 Flex1 225 225 235 +/- 10 => 225 to 245 245 Flex2 15 0 0 +/- 35 => 0 to 35 0 LMP $100 COG 50 50 0 t+1 270 Flex1 215 215 225 +/- 10 => 215 to 235 235 Flex2 30 5 15 +/- 35 => 0 to 50 35 LMP $60 t+2 245 Flex1 225 225 215 +/- 10 => 205 to 225 225 Flex2 45 20 30 +/- 35 => 0 to 65 20 LMP $60 t+3 245 Flex1 235 235 225 +/- 10 => 215 to 235 235 Flex2 45 10 45 +/- 35 => 10 to 80 10 t+4 245 Flex1 235 235 235 +/- 10 => 225 to 245 235 Flex2 75 10 45 +/- 35 => 10 to 80 10 In t, the COG is marginal as it is needed to meet a load increase. It thus sets the MCP. In t+1, the Flex2 unit is marginal and sets the LMP. The COG remains on due to its min run time constraint. In t+2, telemetry shows that Flex1 cannot completely displace Flex2 and thus Flex2 sets the price. In t+3, Flex2 is displaced as the setter of the MCP because it is ramp constrained. Note that Flex2 is still on in the pricing run and thus earning the MCP despite its deviations. MPD/GVB April 4, 2008, page 11