Delmarva Power & Light Company v. Federal Energy Regulatory Commission, Cities of Newark and New Castle, Delaware, Intervenors

BORK, Circuit Judge: Delmarva Power & Light Company petitions for review of a decision of the Federal Energy Regulatory Commission refusing to include the annualized costs of a new generating plant in Delmarva’s test period rate base. Our jurisdiction is based on 16 U.S.C. § 825Z (b) (1982). Delmarva had the burden of establishing that the annualization method it chose resulted in a test-year estimate reasonable when made. The Commission’s Administrative Law Judge (“ALJ”) held that Delmarva had not carried its burden, and the Commission summarily affirmed that aspect of the ALJ’s decision. For the reasons that follow, we conclude that this decision constitutes an unexplained departure from controlling precedent. Accordingly, we vacate and remand to the Commission for explanation of its policy in the matter of selective annualization. I. On April 30, 1980, Delmarva filed a revised tariff increasing its wholesale electric rates in two phases. The Phase I rates were in large part intended to take account of a major expansion in Delmarva’s operating plant — the addition of a new generating unit, Indian River No. 4 (“IR4”), which was expected to begin commercial operation on or about September 1, 1980. The Phase II rates were intended to take account of the addition to plant in service of Delmarva’s 7.41% share of Salem Generation Station Unit No. 2, a new nuclear generating facility which was expected to begin commercial operation in late 1980. Only the Phase I increase is at issue in the case. FERC regulations require that rate filings include information as to the filing utility’s cost of service for two yearly periods. The “Period I” year is the most recent year for which actual data are available. 18 C.F.R. § 35.13(d)(3)(i) (1980). In Delmarva’s case, the Period I data showed Delmarva’s actual cost of service for the year ending September 30, 1979. No dispute over the Period I portion of Delmarva’s filing is before us. This litigation involves only Period II, which the regulation defines as any period of twelve consecutive months after the end of Period I that begins: (A) No earlier than nine months before the date on which the rate schedule change is proposed to become effective; and (B) No later than three months after the date on which the rate schedule change is proposed to become effective. 18 C.F.R. § 35.13(d)(3)(ii) (1980). Because filed rates can become effective only prospectively, the Period II data invariably requires an estimate of cost of service for at least part of the test year selected by the utility. We upheld the validity of this departure from an approach based exclusively on historic costs in American Public Power Association v. FPC, 522 F.2d 142 (D.C.Cir.1975). *1133 Delmarva proposed to make its Phase I rates effective on July 1, 1980. Joint Appendix (“J.A.”) at 1. Under this proposed effective date, it is clear that Delmarva could have chosen a twelve-month period throughout all or nearly all of which IR4 could be expected to be in service as its Period II test year. 1 Delmarva chose not to do so. Instead, Delmarva selected calendar year 1980 as the test year. In testimony filed before the hearing, Delmarva explained this choice on the grounds that “[s]ince Delmarva develops corporate budgets only for the coming calendar year, the forecasted year 1980 is our best available data upon which to set rates.” Testimony of Mr. Gerritson, Tr. at 1269, J.A. at 274. It is uncontested that Delmarva was free to make this choice of a test year under the regulation. Normally, when a utility supplies a test period estimate, the utility must be prepared to show that its estimate was “reasonable when made” in the sense that it represented a reasonable forecast of the test period itself. See, e.g., Villages of Chatham v. FERC, 662 F.2d 23, 29-30 (D.C.Cir.1981). In this case, Delmarva knew that the IR4 unit would not be in service for more than four months during calendar year 1980. Had Delmarva followed ordinary practice, it would therefore have calculated the cost of the IR4 unit for 1980 based on the average of thirteen monthly balances, commencing with the month prior to the test period. Use of this method would have meant that only four-thirteenths of IR4’s estimated cost during IR4’s first year of commercial service would have been included in the test period cost of service — because IR4 was expected to be in operation for only four of the thirteen months in question. Delmarva did not employ the thirteen monthly balance method in preparing its test year estimate, because Delmarva knew that IR4 would almost certainly be in operation during the entire time the proposed rates were in effect. Accordingly, in order to achieve a test year estimate that would more accurately represent cost of service at the time the filed rates were proposed to become effective, Delmarva annualized the cost of IR4 for calendar year 1980. That is, Delmarva assumed that the IR4 facility would be in operation through the test year. Delmarva made this modification to the Period II data in two steps. First, Delmarva removed all IR4-related items from its Period II cost of service. Second, Delmarv'a replaced each removed IR4-related item with the annualized amount for that item. The intervenors in this case urged that FERC’s regulations and decisions embodied a per se rule against annualization. 2 The intervenors thought this point so clear that they urged FERC to reject Delmarva’s filing on this ground — a sanction reserved for filings that “are grossly defective in form, or ‘so patently a nullity as a matter of substantive law, that administrative efficiency and justice are furthered by obviating any docket at the threshold rather than opening a futile docket.’ ” Papago Tribal Utility Authority v. FERC, 628 F.2d 235, 239 (D.C.Cir.), cert. denied, 449 U.S. 1061, 101 S.Ct. 784, 66 L.Ed.2d 604 (1980) (quoting Municipal Light Boards v. FPC, 450 F.2d 1341, 1346 (D.C.Cir.1971), cert. denied, 405 U.S. 989, 92 S.Ct. 1251, 31 L.Ed.2d 455 (1972)). The Commission disagreed, see Order Accepting for Filing and Suspending Revised Tariff and Proposed Increased Rates, Docket No. ER 80-363 (June 30, 1980) (hereinafter “Order Accepting for Filing”), at 3, J.A. at 18, and, in a *1134 prior case arising out of this ratemaking proceeding, we held that FERC’s acceptance of Delmarva’s filing was unreviewable. See Delmarva Power & Light Co. v. FERC, 671 F.2d 587, 593-94 (D.C.Cir.1982). Initially, FERC’s rejection of intervenors’ position was explained on the ground that “the propriety of Delmarva’s ratemaking procedures is an issue to be decided after hearing on the basis of an evidentiary record and is not an appropriate basis for rejection of a filing, such as Delmarva’s, that substantially meets the filing requirements of our regulations.” Order Accepting for Filing at 3, J.A. at 18. Although the Commission refused to reject Delmarva’s rate filing, its order also stated that where phased increases are supported by annualization of the new generating units which come on line during (or after) the test period, the hearing record should also reflect the allocations of the annualized costs with the annualized adjustments of the wholesale billing demands for recovery of the future annualized costs. In other words, the record should reflect the appropriate recognition of the load growth that will occur through the twelve-month period commencing with the commercial operation of the annualized generating facility. Id. The intervenors sought rehearing on the annualization issue. The Commission denied rehearing: We have already stated that annualization of plant for rate base treatment is not per se invalid. Detroit Edison Company, Order on Rehearing, Docket No. ER79-70, issued May 9, 1979. Delmarva will be permitted to present its case in support of its rate base methodology through adjudication. However, as stated in the order of June 30, 1980, the record should reflect recognition of the annualization of the costs as well as recognition of billing demands anticipated during the full first twelve months of the plants’ operation. Delmarva Power & Light Co., 12 F.E.R.C. (CCH) 1161,185, at 61,470 (Aug. 22, 1980), J.A. at 32, 35. In Detroit Edison, which also involved a utility’s attempt to annualize the cost of a new generating plant that was not expected to be in service throughout the test year, the Commission had simply stated that “it is apparent that the annualization proposed by Edison is not proscribed per se by Commission precedent or our current Regulations,” while maintaining that “we do not intend to endorse such modifications to Period II estimates in this or any other proceeding.” Detroit Edison Co., 7 F.E.R.C. (CCH) 1161,145, at 61,226 (May 9, 1979). At the hearing, Delmarva’s witness, Mr. Gerritsen, summarized the annualization method Delmarva employed as follows: “We removed all of Indian River so we can get a total revenue requirement without it and then we put all of Indian River back in on an annualized basis to get a revenue requirement with Indian River and by subtracting the two, being able to quantify the effect of Indian River itself.” J.A. at 324. Thus, for example, with respect to depreciation, Mr. Gerritsen testified that “I annualized depreciation related to major units like Indian River because I was trying to match the test period and annualize it for the commercial operation of those units. I did not annualize depreciation based on the other plant at year end____” Id. at 311. In pre-filed testimony, a Commission staff witness, Ms. Lucas, testified that Delmarva’s annualization should be rejected. Ms. Lucas explained the annualization as follows: “[Delmarva] has adjusted its projected 1980 test period cost support to annualize or to give effect for investment and associated expenses related to the operation of [IR4] for eight additional months.” J.A. at 439. Ms. Lucas’ principal objection to this annualization was that “[Delmarva] has also failed to annualize customer load and sales growth over Period II to reflect condition at the time of the in-service date of the new investments. Certainly if one were to allow the annualization of the operations of new investment, attendant annualizations of sales and load levels must also *1135 be made to reflect system operations during the operation of this investment. To allow annualizations of selective cost of service elements on a piecemeal basis as [Delmarva] proposed without regard to other related elements, is improper because that means that rate base, expenses, revenues and system load are not synchronized or in balance for the test period. All of these test period elements must be in harmony in order to properly test [Delmarva’s] proposed Phase I rate levels.” Id. at 440. Ms. Lucas concluded this portion of her testimony by emphasizing that Delmarva could have chosen a later test year whose commencement “coincided with the expected in-service date of Indian River No. 4. This would have been more representative of average sales, plant, depreciation and other costs when the proposed rates might have [been] allowed to become effective.” Id. At the hearing, Ms. Lucas testified once again. Ms. Lucas stated that she would not “even consider allowing annualization” unless the company proposed an adjustment for customer load and sales growth. J.A. at 588. Ms. Lucas also testified that Delmarva’s filed testimony failed to include, for the first twelve months IR4 was projected to be in operation, (1) the actual load growth; (2) an estimate of billing demands; (3) a projection of metered demands; (4) a projection for coincident peaks; (5) a projection of system peaks; (6) projections of sales on a pro forma basis; and (7) estimates of revenues. Id. at 592-96. Ms. Lucas then suggested that each of these omissions was objectionable: “If the company were going to ... complete their annualization of Indian River No. 4 I believe that it would be necessary to have a complete synchronization of all rate base revenues and expenses.” Id. at 596. On cross-examination by counsel for Cities, the following testimony was elicited: Q: If you annualized every item of revenue and expense to reflect the 12 months from September 1, 1980 through August 31, 1981, would that not be equivalent to use [sic] a test period consisting of 12 months ending August 31, 1981? Ms. Lucas: Yes, it would. Q: Could the company have filed its case using the 12 months ending August 31, 1981 as a test period? Ms. Lucas: The company could have filed its case using a test period ending I believe I stated in my testimony August 27, sometime in August of 1981. Q: Ms. Lucas, if you annualized only a few items related to revenues and expenses of a test period will you achieve a full synchronization of revenues and expenses during the test year? Ms. Lucas: No, you would not. J.A. at 602. The evident implication of this testimony is that the only form of annualization that would achieve a complete, systemwide synchronization of revenues and expenses, while simultaneously including the cost of IR4 for its first year of operation, would be systemwide annualization that was essentially equivalent to a complete test year filing coinciding with IR4’s first year of service. On cross-examination, counsel for Delmarva explored this implication by asking how “the synchronization that you testified to should be accomplished.” J.A. at 605. Ms. Lucas replied that “there should be a matching of all rate base expense, revenues and system load costs during the test period.” Id. Counsel asked how that could be done with a 1980 test period and a September 1980 in-service date for IR4. Ms. Lucas said “I cannot tell you how to go about it. I would have to say the only thing that I would recommend for a company who knew in advance that they had an investment that was going into service on that date and they were going to file for some type of rate increase and they want to include this investment in rate base, I would suggest that they provide a cost of service or test period that would coincide with the proposed in-service date of that unit.” Id. at 606. Asked whether this meant she “would support no annualization,” Ms. Lucas answered, “I think it is necessary to look at the Commission’s reg *1136 illations.” Id. When pressed, Ms. Lucas denied that she was ruling out annualization. Id. But when asked “[l]et’s take your hypothesis that a company must file a test period that begins on or before — on or after the expected in-service date of a new generating unit, then isn’t it true that the question of annualization just does not arise?” Ms. Lucas replied, “[tjhat would be correct. The investment would already be in rate base.” Id. at 606-07. We now turn to the opinion of the Commission’s Administrative Law Judge. The AU reviewed the Commission’s opinions and concluded that “Delmarva is correct that annualization is not prohibited. Yet the strong Commission precedent ... clearly indicates a strong policy against such a procedure.” J.A. at 86. In the AU’s view, the premise of this policy was the judgment “that a particular expense or rate base item may affect more than a few expense or rate base accounts and may affect revenue accounts and allocation factors. Only through a complete and consistent set of test year projections will the effect and interrelationship of all items of the Company’s cost of service and rate base be properly reflected in the rates charged.” Id. The AU read the Commission’s statements in the orders setting the case for hearing and clarifying the hearing order as simply restating this principle. The AU then rejected Delmarva’s contention that test year projections must be adjusted where they would yield unreasonable results, on the grounds that this contention missed the mark: “Delmarva seeks to change unquestionably accurate projections concerning the test year in favor of post-test year data.” J.A. at 86. In other words, the AU took the test year as the benchmark for assessing the accuracy of the projections, not the period during which the rates were proposed to be in effect. The AU next stated his understanding of the circumstances in which annualization is not prohibited: “Where such an annualizing adjustment is made, it must include recognition of the load growth and any other factors which should be considered in order to achieve a proper matching of expenses and revenues.” J.A. at 86. He summarized the testimony of Ms. Lucas and Munies’ witness that Delmarva’s annualization failed to achieve this matching. The AU plainly agreed with these witnesses that the annualization adjustment, standing alone as part of Delmarva’s filing, did not achieve the requisite matching. Nonetheless, the AU appears to have believed that Delmarva’s proposed annualization should be accepted if Delmarva could “prove that all other test year estimates which were not annualized were not affected by the new generating plant.” J.A. at 87. He concluded, however, that Delmarva had not met this burden. In order to understand the nature of this burden, it is necessary to examine the AU’s response to Delmarva’s contention that “its ‘annualization includes estimates of sales from the first twelve months of commercial operation’ for the new unit and also includes estimates of increased earnings and actual fuel costs.” J.A. at 87 (emphasis added). The AU did not deny that Delmarva had made this showing— that, in other words, Delmarva had annualized all items that, at least according to Delmarva, were directly related to the IR4 facility. Rather, the AU viewed that showing as insufficient, because it did not address the possible impact of IR4 on systemwide test year estimates. Thus, the AU said that “the burden is on the Company to show that there was no revenue, load and sales growth to be annualized.” Id. He rejected Delmarva’s evidence that (1) increases in plant capacity do not necessarily lead to load growth, (2) 1980 sales were lower than 1979 sales, and (3) the peak load during IR4’s first year in operation was lower than projected, as inadequate to sustain Delmarva’s burden. What the AU required was direct proof that, for example, “there was no appreciable increase in load.” Id. Since IR4’s first year of operation ended nearly one year before the hearing commenced, the AU thought Delmarva had ample opportunity to make this showing and similar showings as to other test *1137 year estimates. Accordingly, the AU rejected the proposed annualization of IR4 despite his recognition that “including the annual costs of Indian River No. 4 will make those particular costs more representative of the costs actually incurred during the effective period of the rates.” Id. (emphasis added). II. Delmarva asks us to reverse the AU’s decision on the grounds that Delmarva established that its proposed annualization was reasonable and consistent with the test year principié. In the alternative, counsel for Delmarva stated at oral argument that Delmarva would accept a remand to give it an opportunity to meet the burden of proof the AU imposed. Counsel for the Commission stated that this course would not be acceptable to the Commission because Delmarva had adequate notice of its burden and ample opportunity to meet it. We asked the parties to submit supplemental filings on the notice issue, but have concluded that our decision should not turn on whether Delmarva had notice of its burden of proof. Section 825Z (b), which provides for judicial review of final orders issued by the Commission, specifically states that “[n]o objection to the order of the Commission shall be considered by the court unless such objection shall have been urged before the Commission in the application for rehearing unless there is reasonable ground for failure so to do.” The purpose of section 825Z (b) is “to insure that the Commission has an opportunity to deal with any difficulties presented by its action before the reviewing court intervenes.” Villages of Chatham v. FERC, 662 F.2d 23, 30 (D.C.Cir.1981) (quoting Rhode Island Consumers’ Council v. FPC, 504 F.2d 203, 212 (D.C.Cir.1974)). We have accordingly interpreted section 825Z (b) to mean that “any argument brought clearly to the attention of the Commission by the party’s petition for rehearing has been preserved for review.” Id. See also Belco Petroleum Corp. v. FERC, 589 F.2d 680, 683 (D.C.Cir.1978). As this court noted in Kansas Cities v. FERC, 723 F.2d 82, 85 (D.C.Cir.1983), compliance with this provision “is a prerequisite to judicial review.” See FPC v. Colorado Interstate Gas Co., 348 U.S. 492, 498-99 & n. 4, 75 S.Ct. 467, 471-72, & n. 4, 99 L.Ed. 583 (1955) (construing an identical provision in Natural Gas Act as precluding judicial review when objection not preserved before agency). Although FERC does not argue that section 825Z (b) precludes resolution of Delmarva’s notice claim, we have therefore carefully examined the record to determine whether Delmarva’s claim that it had no notice of the evidentiary burden the AU imposed on it, and should therefore be given an opportunity to make the required showing, was raised before the Commission. We think not. In its Brief on Exceptions to the AU’s decision, one of Delmarva’s principal objections was that “[t]he Judge misapplied the Commission’s Orders of June 30 and August 22, 1980.” J.A. at 104. Delmarva argued that the AU was mistaken in thinking that the passage referring to annualization concerned Phase I rates, id. at 105 n. *; that he was mistaken in concluding that the Commission meant to require evidence of actual rather than projected billing demands, id. at 104; and that he was mistaken in assuming that Delmarva had a duty “to introduce other evidence of billing demands in addition to its filed data.” Id. at 105. In this connection, Delmarva said, “it must be emphasized that nowhere in either the June 30 or August 22, 1980 Orders did the Commission specify that Delmarva alone had the duty alleged by Staff and Munies to introduce specific data into the record.” Id. at 106. Clearly, then, Delmarva raised a claim that the AU had failed to follow the Commission’s instructions, and erroneously imposed a heavy evidentiary burden on Delmarva in the belief that the Commission had already so ruled. This is not a true notice claim. Delmarva did not ask the Commission to give it an opportunity to meet the burden imposed by the AU. Lack of notice of a burden is an objection quite distinct from *1138 the asserted impropriety of the burden itself. Thus, it is clear that Delmarva could have argued in the alternative that even if the ALJ had imposed the proper burden of proof the Commission should grant it an adequate opportunity to satisfy that burden. Delmarva’s petition for rehearing did not clearly put to the Commission the contention that Delmarva was denied an opportunity to meet the burden of proof the ALJ imposed. We therefore do not reach this claim. 3 Delmarva’s central contention, however, is properly before us. That contention, in a nutshell, is that the ALJ’s decision rejecting selective annualization of IR4 is contrary to the purpose of the test year regulation and unsupported by substantial evidence, and that the burden of proof the ALJ assigned Delmarva was inconsistent both with the Commission’s earlier orders in this case and with this court’s decisions on the subject of evidentiary burdens in test year disputes. We now take up this contention. III. When the Commission adopted its test year regulations in 1973, it explained that “[t]he effect of the Period I and Period II data will be to create a test period which in most cases will represent costs six months prior to and six months after a rate change would become effective.” Filing of Electric Service Tariff Changes, Order No. 487, 50 F.P.C. 125, 127 (1973). One purpose of this departure from historic cost ratemaking, then, was to make it more likely that rates would accurately reflect the utility’s actual cost of service at the time when the rates went into effect. At the same time, however, the Commission also emphasized the administrative convenience to be expected from use of Period II cost estimates. Id. at 127. The Commission also gave some weight to the argument that rates based in part on estimates of future costs might be speculative and manipulable: it stated unequivocally that it “will not approve rates based on unsubstantiated cost estimations. The burden [is] on such companies to establish the validity and accuracy for each of their cost estimates.” Id. Since upholding the validity of the revised test year regulations in American Public Power, 522 F.2d at 145, this court has decided a number of cases involving disputes over the reasonableness of Period II test year estimates. 4 Generally, these cases have come to us as challenges by wholesale customers who are seeking to show that the utility’s estimates were unreasonable in light of actual data on the period for which the estimate was offered. See, e.g., Boroughs of Ellwood City v. FERC, 731 F.2d 959, 965-67 (D.C.Cir.1984) (challenge to utility’s estimate of revenues from sales of excess reserve capacity based on actual data for the test year); Villages of Chatham, 662 F.2d at 30-35 (challenges to utility’s estimates of customers’ coinci *1139 dent peak demand, of utility’s purchased power credits, and of utility’s tax rate, all based on actual data for the test year); Indiana & Michigan Municipal Distributors Association v. FERC, 659 F.2d 1193, 1197-98 (D.C.Cir.1981) (challenge to utility’s estimate of wholesale short-term sales based on actual data for the test year). These cases have established a framework for analysis of the types of test-year questions we have hitherto faced: The utility must demonstrate that its estimates were reasonable when made, either by explaining the chain of reasoning that it employed to arrive at its projections, or by comparing its estimates with actual data, and so establishing their accuracy. If the utility proves that its estimates were reasonable when made even though at variance with the actual results, the Commission may nonetheless credit those estimates, a decision further substantiated by evidence indicating that the actual figures are based upon unusual or unique occurrences. And of course the Commission is obligated to consider contrary evidence introduced by customers or other parties, as well as the magnitude of the disparity between expectation and reality. Villages of Chatham, 662 F.2d at 29-30. We think that the burden of proof the AU imposed on Delmarva in this case constitutes an unexplained departure from this framework. That error, in turn, traces back to the Commission’s failure to explain when and why selective annualization is consistent with the test year regulation. As previously noted, the presupposition of the rule that a test year estimate be reasonable when made is that the estimate is an effort to predict actual experience for the test year chosen. If the test year estimate is shown to be reasonable, the test year estimate, averaged in with the Period I actual data, yields a new rate base from which the utility’s new revenue requirement, and hence its rates, are derived. See 18 C.F.R. § 35.13(a)(iii) (1980). The fact that actual costs during the period in which the rates are actually in effect may differ from actual costs during the period for which the test year estimate is furnished thus provides no basis for customers to challenge the filed rates. For example, if the test year is calendar year 1981 but the rates are not proposed to become effective until January 1, 1982, there is no indication in the cases or in the Commission’s regulations that customers could challenge the 1981 estimates as unreasonable on the basis of actual data for 1982. Nor, continuing the same example, have we found any suggestion that a utility could rebut a showing that its test year estimate was inordinately and foreseeably different from actual data for the test year by establishing that the 1981 estimate accurately predicted actual data for 1982. Had the Commission adhered to this understanding of the test year’s function, we find it hard to see how it could have stated that selective annualization is not per se prohibited by the test year regulation. The whole premise of Delmarva’s annualization of IR4 is that this modification is necessary to predict actual experience for the first year for which the rates are proposed to be in effect. Thus, Delmarva proposed a test year based primarily on data for calendar year 1980, but then introduced selective annualization on the grounds that if it did not do so the test year estimate would not accurately reflect cost of service for a different period — the first year during which the rates were proposed to be in effect. We think it apparent, then, that the Commission could have taken the position that Delmarva’s proposal was flatly inconsistent with the test year regulation, because Delmarva’s composite test year was not subject to comparison with actual systemwide results for the period the test year purported to represent. That is to say, we think this understanding of the test year regulation is a reasonable one that it is within the Commission’s discretion to adopt. 5 *1140 It is clear, however, that the Commission has declined to adopt such a per se rule against selective annualization. Yet the Commission has also refrained from endorsing selective annualization, and it has not explained how a utility should proceed in order to show that test period data utilizing selective annualization are reasonable, or how parties objecting to the utility’s filing should seek to establish otherwise. We find it impossible to discern in the pronouncements of the Commission itself, either in Detroit Edison or in its initial orders in this proceeding, a test for determining when selective annualization will be allowed. This taciturnity is not helpful to the parties or to a reviewing court. Indeed, the Commission’s statements in its initial orders concerning what the record should show are susceptible of at least three different interpretations. The first, which Delmarva urges upon us, is that those instructions “were addressed to the possibility that operation of the new facility might cause growth in customer load extensive enough to warrant attention in the allocation process.” Supplemental Brief of Delmarva at 11. We think it unlikely that Delmarva is right, because we see no reason why the Commission would be concerned that IR4 would cause load growth that would distort the allocation process and yet not be concerned that the same load growth would impair the reliability of Delmarva’s hybrid test year estimate. The second interpretation is that the Commission meant that the addition of IR4 might cause load growth and that Delmarva should therefore include in its filing an estimate of load growth for the first year in which IR4 was expected to be in service. The third interpretation is that Delmarva was required to supply actual data on load growth for IR4’s first year in operation. Under either of these interpretations, two further ambiguities would remain: did the Commission intend that Delmarva’s annualization be rejected if there was any load growth, or merely that an adjustment be made to Delmarva’s test year estimate to reflect that load growth? And did the *1141 Commission intend that any load growth be presumed attributable to IR4, or did it intend that Delmarva could show that any load growth, estimated or actual, was attributable to other causes? We need not choose between these two interpretations, because it is clear that the ALJ’s decision does not rest on a conclusion that Delmarva did not comply with the Commission’s instructions, read in either of these alternative ways. It is true that the AU faults Delmarva for failing to “show that there was no load growth associated with the additional capacity,” J.A. at 87, but that is only one of several deficiencies the AU perceived in Delmarva’s case for annualization. As the AU’s recitation of the testimony opposing selective annualization confirms, the AU thought Delmarva’s filing, standing alone, should be rejected because Delmarva did not annualize systemwide revenue items. We see no way in which the Commission’s instructions concerning what the record should show can be read to require such an across-the-board showing. Consequently, the question becomes whether such an across-the-board showing is consistent with the Commission’s more general statement that selective annualization is not per se prohibited, and with the general burden of proof a utility bears in connection with its test year estimates. We think both questions must be answered in the negative. An across-the-board annualization rule is virtually indistinguishable from a per se rule against selective annualization. As Ms. Lucas admitted, to require the filing utility to annualize every item in its filing for its entire system is equivalent to requiring the utility to choose a later test year — one that coincides with the commencement in operation of the new plant. Had the AU simply adopted Ms. Lucas’ position, therefore, our task as a reviewing court would be straight-forward. The across-the-board annualization requirement appears entirely inconsistent with the Commission’s pronouncements on annualization of new plant in rate base. If the Commission meant that it would allow such annualization only when carried out for all revenue and expense items in a utility’s entire system, it certainly did not say so. 6 Accordingly, we would remand to the Commission for an explanation of its position. Our task is complicated, however, by the fact that the AU suggested that the basis for his decision was Delmarva’s failure to meet its burden of proof. Yet, as we will show, the AU’s treatment of Delmarva’s burden of proof is essentially a restatement of the per se approach in evidentiary terms. Accordingly, we conclude that remand is indeed required. The AU appears to have decided that even though Delmarva’s filing was defective because Delmarva had not annualized revenue items for its entire system, he would accept selective annualization if Delmarva showed that “there was no revenue, load and sales growth to be annualized.” J.A. at 87. But that means Delmarva could have succeeded only if its actual system-wide revenue, load and sales figures for the first year in which IR4 was in operation did not exceed Delmarva’s modified figures for the 1980 test year. Under the AU’s approach, therefore, Delmarva was required to show that its 1980 test year estimates (including IR4 annualization) unerringly predicted experience for a different, *1142 later test year. 7 The AU did not suggest that he would have accepted a showing that the 1980 test year estimate was a reasonable forecast of systemwide experience for the first year during which IR4 was in service. Nor did the AU suggest any other way in which Delmarva could prove the negative proposition that the addition of IR4 to Delmarva’s system had no effects on any of the items in Delmarva’s rate base that were not already reflected in Delmarva’s annualization of IR4-related items. In short, we think the AU imposed a burden of proof on Delmarva that requires that an estimate be flawless in retrospect rather than reasonable when made. 8 We know of no precedent for such a requirement. Beyond that, it is clear that the AU imposed this burden because he believed that Delmarva’s methodology was not reasonable — and that the basis for his belief was not factual evidence that Delmarva’s filing understated systemwide revenue items but rather a generalized suspicion that this possibility exists whenever selective annualization is employed. That is why the AU demanded proof “that all other test year estimates which were not annualized were not affected by the new generating plant.” J.A. at 87. We conclude, therefore, that the AU’s solution to the dilemma in which the Commission’s inconclusive pronouncements on selective annualization placed him is itself tantamount to a per se rule against selective annualization. No utility could conceivably afford to file estimated data for test year x, including selective annualization of items related to a facility that will commence service in year x + 1, in the hope that it would eventually be able to prove that actual data for year x + 1 correspond perfectly to estimated data for year x. Hence, every utility facing the addition of a new plant to its rate base will choose the latest Period II test year permitted by the Commission’s regulation. That is what Ms. Lucas advocated, and Delmarva’s failure to do so prompted the AU to conclude that the annualization issue was a problem Delmarva brought on itself. The AU’s solution, were we to affirm, would guarantee that no utility will propose selective annualization in the future. We think affirmance under these circumstances is clearly inappropriate. The parties have treated the Commission’s instructions in its initial orders as binding. By summarily affirming, the Commission has precluded any argument that it reconsidered those original instructions and concluded that they were ill-advised and that a per se rule against selective annualization is in fact appropriate. Those instructions were lamentably unclear, in every respect but one: the Commission stated that it has no such per se rule. The AU, relying on testimony from a Commission staff witness that quite clearly advocated a per se approach, interpreted those instructions from the Commission in a manner that, albeit somewhat circuitously, reaches the same result as a per se rule. Under these circumstances, we have no choice but to characterize this case as one in which the Commission has taken inconsistent positions without ever explaining either of those positions, let alone explaining the change *1143 from the former position to the latter. That is arbitrary and capricious. 9 However, because, as we have said, there is a real possibility that, put to a choice, the Commission would indeed adopt a per se rule against selective annualization, and because the test year regulation would appear to provide ample basis for doing so, we cannot simply reverse and direct the Commission to allow Delmarva’s selective annualization. The Chenery principle requires that we remand to the Commission to consider whether to adopt this alternative ground for the AU’s decision. See SEC v. Chenery Corp., 318 U.S. 80, 63 S.Ct. 454, 87 L.Ed. 626 (1943). Of course, the Commission may have a policy, akin to law of the case, that would preclude it from applying a per se rule it rejected in its initial orders upon reconsideration at a later stage in the same proceedings. We express no opinion on that point. Nor do we rule out the possibility that the Commission may conclude that its policy with respect to selective annualization falls short of a per se rule against the practice but nonetheless would not allow Delmarva to annualize IR4-related items in this case. Any such decision, however, must be accompanied by an adequate and reasoned explanation of the Commission’s policy and a showing that the result reached by application of that policy to Delmarva rests on substantial evidence. Finally, although we have concluded that we cannot consider Delmarva’s claim that it had no notice of the novel evidentiary burden the AU imposed and should be given an opportunity to make the showing he required, nothing in our decision bars the Commission, should it conclude that some such special burden should be borne by a utility proposing selective annualization, from considering whether Delmarva should be afforded such an opportunity. We vacate and remand to the Commission for explanation of its policy in the matter of selective annualization. It is so ordered. Footnotes [1] The Commission later recharacterized Delmarva’s filing as effectively requesting an effective date for Phase I "coincident with the in-service date[]” of the IR4 unit. Docket No. ER 80-363, Order Clarifying Prior Order and Denying Rehearing at 4 (Aug. 22, 1980), J.A. at 35. Thus, the proposed effective date for Phase I as determined by the Commission would appear to be October 1, 1980, when IR4 commenced commercial operation. [2] The Cities of Newark and New Castle, Delaware ("Cities”), have intervened in this case. Cities also intervened before the Commission, as did several other Delaware municipalities and a group of electric cooperatives. [3] Hatch v. FERC, 654 F.2d 825 (D.C.Cir.1981), closely resembles the present case, but one salient difference distinguishes them. In Hatch, the Commission "denied Petitioner's assertion that the standard for obtaining authorization [to hold interlocking directorships in certain corporations] had in fact been changed without benefit of either rule-making proceedings or advance notice to Petitioner of what he must prove in the evidentiary hearing.” Id. at 830 (emphasis added). Because both the claim of agency inconsistency and the notice claim had been presented to the Commission, we addressed both issues. We found that FERC had indeed changed the relevant standard without explanation, and that petitioner had not received sufficient notice of the change "to warrant denial of an opportunity to supplement the record to meet the new standard.” Id. at 837. Here, Delmarva failed to request such an opportunity from the Commission. [4] In American Public Power, 522 F.2d at 145, this court rejected the argument that "rates based upon estimates of future costs are too speculative to be 'just and reasonable’; and that the establishment of such rates would be based upon estimates made by the utilities and could therefore be manipulated by them to their financial advantage.” The American Public Power court found that Congress had not "specifically endorsed] historic test year ratemaking or any other technique of ratemaking," and had given the Commission broad discretion in selecting an appropriate methodology for rate-making. Id. at 146. [5] Indeed, precisely the position we have just outlined was adopted by a Commission ALJ in Indiana & Michigan Electric Co., 11 F.E.R.C. (CCH) ¶ 63,037 (June 18, 1980). The ALJ held *1140 that the utility’s selective annualization of a new generating unit "is an impermissible deviation from the methodology prescribed in this Commission’s Regulations.” Id. at 65,222. Since it was undisputed that the utility "did not expect the [new facility] to be in service for the entire Period II test year,” the AU found that the test year estimate was “unreasonable and excessive when made.” Id. at 65,223. Surprisingly, the AU adopted this position after the Commission had reversed itself in Detroit Edison and unequivocally asserted that the test year regulation does not per se prohibit selective annualization of new plant in rate base. The utility invoked Detroit Edison, but to no avail. The AU’s response was that in Detroit Edison the Commission merely accepted "annualized figures for filing,” without endorsing such estimates in the absence of some showing that they were justified. 11 F.E.R.C. (CCH) at 65,-223. The AU then stated that ”[t]he record in this proceeding contains no evidence concerning annualization which would justify a deviation from the Commission’s prescribed test year methodology.” Id. The AU did not explain what form such evidence should take. We think it obvious that to distinguish Detroit Edison on these grounds is to ignore that case. However, the AU in Indiana & Michigan Electric went on to say that even if annualization were permissible, it would only be allowed when "accompanied by the adjustment of those related line items necessary to achieve synchronization of revenues and expenses.” 11 F.E.R.C. (CCH) at 65,223. Because the utility had conceded that its selective annualization did not "result in a complete synchronization of revenues and expenses," id. at 65,223-24, the ALJ thought its annualization would be forbidden for this reason as well. This interpretation of the synchronicity requirement holds that all revenue and expense items for the utility’s entire system must be synchronized. The Commission’s staff witness, Ms. Lucas, took the same position in this case. As we explain infra, we find this interpretation equivalent to a per se rule against selective annualization, and consequently we find this interpretation extremely difficult to reconcile with the Commission’s claim that it has no per se rule against annualization of new plant in rate base. The other interpretation is illustrated by Pennsylvania Power Co., 9 F.E.R.C. (CCH) ¶ 63,041 (Nov. 7, 1979), where the AU rejected a utility’s inclusion of a full year’s depreciation expense on a new plant that was only in service for part of the test year because revenues for that unit were not annualized while depreciation was. That is the understanding on which Delmarva proceeded to annualize only IR4-related items. The Commission has not endorsed that interpretation either. [6] In Detroit Edison, the Commission originally rejected the utility’s selective annualization. The record before the Commission clearly established that the annualization of Detroit Edison’s Greenwood plant was selective rather than across-the-board. As the Commission stated, "[ajccording to Edison, its Period II cost of service annualizes the Greenwood investment and costs in order to avoid ‘distortion’ in the test period data. Edison further states that it has also annualized related cost of service adjustments and has made offsetting adjustments such as a reduction in the Period II purchased power expense.” 6 F.E.R.C. (CCH) ¶ 61,218, at 61,548 (Mar. 9, 1979). Hence we see no possibility that the Commission was unaware, when it stated that "the annualization proposed by Edison is not proscribed per se by Commission precedent or our current Regulations," 7 F.E.R.C. (CCH) ¶ 61,145, at 61,226 (May 9, 1979), that the proposed annualization was a selective rather than an across-the-board adjustment. [7] Under the across-the-board annualization approach advocated by Ms. Lucas, Delmarva would in effect have been required to use IR4’s first year in service as its test year, because Delmarva would have been required to furnish annualized estimates for each item in its filing. Had Delmarva done so, its annualized estimates could have been challenged by intervenors or by the Commission's staff, but the challenging parties would have been required to introduce some evidence that Delmarva’s estimates were not reasonable when made — that is, evidence that actual data for IR4’s first year in service were substantially at variance with Delmarva’s estimates. [8] The AU might have accepted proof that any deviation between the test year estimate and actual data for the first year in which IR4 was in service did not favor Delmarva. That would still mean that the ALJ required at least an asymmetrical version of perfection; it does not mean that the ALJ was employing the normal reasonableness standard. [9] It is a familiar tenet of administrative law that "[r]eview of the reasonableness of an administrative adjudication includes consideration of the administrator’s consistency in deciding similar cases.” Dep't of Treasury v. FLRA, 707 F.2d 574, 581 n. 25 (D.C.Cir.1983) (citation omitted). While it is generally true that “an agency is free to alter its past rulings and practices even in an adjudicatory setting ... it is equally settled that an agency must provide a reasoned explanation for any failure to adhere to its own precedents.” Hatch, 654 F.2d at 834 (citations omitted). Treating FERC’s pronouncements in Detroit Edison and in its initial orders in this case as "precedents,” this is plainly a case in which FERC has "glosse[d] over or swerve[d] from prior precedents without discussion ... [and] crosse[d] the line from the tolerably terse to the intolerably mute.” Greater Boston Television Corp. v. FCC, 444 F.2d 841, 852 (D.C.Cir.1970), cert. denied, 403 U.S. 923, 91 S.Ct. 2229, 29 L.Ed.2d 701 (1971). Ordinarily, of course, the "precedents” that an agency must explain its failure to adhere to are themselves reviewable final decisions contrary to the decision under review, whether handed down by the agency prior to or subsequent to the decision under review. See Dep't of Treasury, 707 F.2d at 581 n. 25. Detroit Edison was not such a reviewable final decision — that was an order accepting the utility’s filing, and preserving the annualization issue for litigation at the hearing. The same characterization applies to the initial orders in the present case. We decline to adopt a flat rule that an agency’s statements in preliminary orders in a ratemaking proceeding can never be treated as precedent for purposes of determining whether the agency decision under review constitutes an unexplained change of position. We also decline to adopt the opposite rule. Instead, we base our conclusion that Commission precedent holds that there is no per se rule against selective annualization on three factors: (1) the parties have treated the Commission’s initial orders as controlling law in this case; (2) the Commission in its initial orders treated Detroit Edison as binding precedent; and (3) the issue as to which we are treating FERC's prior statements as precedent is a question of law — whether selective annualization is consistent with the test year regulation. In relying on the first two factors we are simply taking this case as the parties have framed it; in relying on the third factor we are recognizing that the Commission’s statements appear to have adopted an interpretation of the test year regulations without preserving that issue for litigation at the hearing before the AU. Under these circumstances, we think that the concern that animates the consistency requirement — ensuring "the evenhanded declaration and application of the law by an administrative authority," Dep’t of Treasury, 707 F.2d at 581 n. 25 — is present here much as it is in the more common situation where an agency has reached inconsistent results, in each instance after full adjudication.