In this case, municipal customers of Connecticut Light and Power Company ("Connecticut Light" or "Company") challenge a decision of the Federal Energy Regulatory Commission ("FERC" or "Commission") to approve in major part a rate design submitted by Connecticut Light. The municipalities are members of the Connecticut Municipal Electric & Gas Association ("CMEGA"); they include the Third Taxing District of Norwalk, a customer drawing its full power requirements from Connecticut Light, and the Second Taxing District of Norwalk and the Town of Wallingford, both customers who receive only a portion of their power requirements from the Company. The CMEGA members are understandably interested in paying low rates for their service from Connecticut Light, and in being able to shift to even cheaper sources of wholesale power when possible. The Company, which has large amounts of underutilized generating capacity, is just as understandably concerned to ensure that the municipalities bear their fair share of the costs of the facilities on which they rely for their power.

This proceeding is the fourth in a series of efforts by Connecticut Light to tailor rates to costs of service by the use of a rate design that charges rates for power based on whether it was consumed at periods of highest demand on the system. CMEGA objects specifically to the method used by the Company to calculate rates charged partial requirements customers for the costs borne by the Company in maintaining the capacity to serve their power demands. CMEGA contends that approval of the rate design is a sharp departure from FERC's treatment of the full requirements customers in this case, FERC's treatment of previous rate designs by Connecticut Light, and FERC policy on rate design more generally. CMEGA also challenges FERC's refusal to order a refund for the full requirements customers, who were not made subject to some features of the rate design applied to the partial requirements customers. Finding substantial record support for FERC's decision on the rate design, we affirm.

I. THE CHALLENGED RATE DESIGN AND ITS HISTORY

The general aim of rate design is to enable a utility to recover allowable revenues by charging customers utility rates that reflect the costs of providing service. As is standard, Connecticut Light began the process of determining the rate to be charged its wholesale municipal customers by distinguishing between "demand costs" and "energy costs."1 "Demand costs" reflect the costs to the Company of maintaining the facilities needed to meet its customers' requirements. "Energy costs" are the costs incurred in running the facilities to generate each unit of power actually produced. Individual customers' bills are the sum of a demand charge, calculated to reflect the customer's share of demand costs, and an energy charge, calculated to reflect the costs of producing the power used by the customer.

In the rate design at issue here, Connecticut Light figured demand charges in two stages. First, the Company allocated total demand costs among customer classes by utilizing the 12-month coincident peak (12-CP) method. Under this method, demand costs are allocated by taking the hour of highest total usage (the coincident peak) during each of the preceding twelve months, determining the percentage of peak usage drawn by each customer class during each of the twelve months, and averaging the resulting percentages for each customer class. For example, if a customer class such as the CMEGA municipalities was responsible for 30% of usage at coincident peak during the three summer and three winter months, and 20% of usage at coincident peak during the remaining six months, the class would be allocated 25% of demand costs. Because a public utility is expected to maintain the capacity to meet its customers' needs, a customer class's highest "contribution" to coincident peak will reflect the added costs incurred by the utility in maintaining the capacity to meet the needs of that class, at least, if usage by other customer classes remains constant. For example, if municipal wholesale customers draw a higher percentage of power used at coincident peak during the summer months because of the heavy use of residential air-conditioning, and if other users draw relatively steady amounts of power throughout the year, the municipalities' contribution to coincident peak during the summer represents the increased costs incurred by the Company in meeting their demands. Because it averages monthly contributions to coincident peak for an entire year, therefore, the 12-CP method best reflects real costs imposed on the system by customer classes when the classes tend to use fairly steady amounts of power throughout the year. It least reflects actual costs when a customer class imposes particularly heavy seasonal demands, as in the air-conditioning example.2

Once demand costs were allocated to customer classes, the Company then computed the portion of the class's allocation to be borne by each class member. In the rate design at issue here, Connecticut Light calculated individual demand charges on the basis of a stratified rate, applied to the customer's "billing demand": the maximum clock hour of kilowatt demand supplied by the Company to the customer during the previous twelve months. The rate was "stratified" in that the customer's billing demand was subdivided into two tiers: usage at "peak" and at "off-peak" hours. The Company charged a higher demand charge for usage during periods of off-peak demand, and a lower demand charge for peak use, because the newer nuclear plant used to supply off-peak capacity is the more expensive portion of the Company's capital facilities. The Company's method of billing a demand charge based on the customer's full amount of highest demand during the past twelve months is called a "twelve-month, 100% billing demand ratchet." The "ratchet" effect is that a customer which in one month imposes high demands on the Company, for whatever reason, will not slip back to paying demand costs based on a lower amount of usage until twelve months have passed, no matter how severely the customer curtails energy use.3

In addition to the demand charge, the rate at issue here imposed an energy charge on Company customers. The charge was based on the number of kilowatt hours actually drawn by the customer from Connecticut Light each month. Like the demand charge, the proposed energy charge was stratified, but with a higher rate charged for usage during peak periods and a lower rate charged for off-peak usage. This stratification too was designed to track costs of service, because the Company must employ less fuel-efficient plant to generate the larger amounts of power consumed at periods of peak usage.

The rate design at issue here, R-4,4 was the fourth in a series of stratified rates submitted by Connecticut Light to the Commission. Although the stratified rates were intended to tailor wholesale rates5 more closely to the costs of providing service, they initially met with little success before the Commission. The first, R-1, filed in 1972, proposed a rate stratified into four levels, with different portions of a customer's load assigned to the different strata but without a distinction between demand and energy charges. R-1 imposed the highest charges on partial requirements customers, by assigning initial portions of their use to the level at which the charge was highest, and employed a 100% ratchet based on the customer's total demands for power, from whatever source. While noting that "the concept of tailoring rates to the service provided is worthy in the abstract," the Commission rejected the stratification proposed in R-1 because the method by which it had been developed had not been shown to track actual costs of service. The Connecticut Light & Power Co., Opinion No. 761, 55 F.P.C. 1986, 1999 (1976). The Commission also rejected the total demand ratchet in the form proposed. The Company had defended including a customer's total demands for power-including power generated by the customer itself-in the ratchet by arguing that it needed to maintain system capacity to meet its wholesale customers' entire potential demand, in case of outages by alternate suppliers at times of peak demand. The Commission, however, found that the ratchet in this form imposed a burden on customers wholly beyond the actual costs of serving them. Id. at 1996.6

The R-1 proceedings, initiated in 1972, were not completed until 1976. In 1974, after the initial decision by the ALJ had rejected the stratified rate proposed in R-1, Connecticut Light filed the second rate in the series, R-2, which also proposed a stratified rate coupled with a demand ratchet. R-2, however, was importantly different from R-1. The R-2 rate was subdivided into a demand charge and an energy charge, each in turn stratified into only two tiers, with one charge for peak usage and the other for off-peak usage. The ratchet in R-2 was based solely on the highest amount of power actually drawn by the customer from Connecticut Light during the preceding twelve months, not on the customer's total power demand, however supplied. In December, 1975, R-2 was superseded by Connecticut Light's filing of R-3, a stratified rate similar in structure to R-2. See The Connecticut Light and Power Co., Docket No. ER76-320, Initial Decision (Oct. 16, 1978). R-2 was eventually settled during the R-3 proceedings.

In the R-3 proceeding, both the stratified rate and the demand ratchet were rejected. The ALJ rejected the stratification as proposed because he found that the Company had not made an adequate showing of how the proposed rate differentials tracked actual costs of service.7 He rejected the proposed 100% ratchet in favor of an 80% ratchet because of concern that it might result in overcollections when combined with the 12-CP method of allocating demand among customer classes.8 FERC affirmed without comment. The Connecticut Light & Power Co., Opinion No. 103, 13 FERC P 61,155 (1980).

R-4, the proceeding at issue here, intermeshed with the proceeding in R-3. R-4, filed July 31, 1978, incorporated the same two-tier stratification structure and 100% billing demand ratchet as R-3 and R-2. The only differences between R-3 and R-4 of importance here concern the dividing line between power use classified as "peak" and "off-peak" and the amount of reserve capacity postulated for the purposes of calculating the actual peak and off-peak demand charge rates. The ALJ issued his initial decision in R-4 in September, 1980, some two months before FERC's final order in R-3. In R-4, however, the ALJ approved the stratified rate design, finding that the Company had shown that the design matched costs of service with charges for service in a technically accurate manner. Connecticut Light & Power Co., Initial Decision, No. ER78-517 (Sept. 9, 1980); Joint Appendix (J.A.) 642. The ALJ rejected the 100% ratchet for wholesale customers drawing all of their power from Connecticut Light, because he found that the Company had not made the showing requisite to override the ratchet's potential for unfairness. Id. at 668. The ALJ did not, however, consider the application of the ratchet to partial requirements customers, because he concluded that the issue had not been raised. Id. at 672 n.37. FERC affirmed, but on the ground that application of the ratchet to partial requirements customers was justified by their ability to utilize alternate power sources in order to limit their peak demand from Connecticut Light. The Connecticut Light & Power Co., Opinion No. 114 (Feb. 19, 1981), J.A. 834. On all other issues, FERC adopted the reasoning of the ALJ.

The CMEGA municipalities petitioned for rehearing and Connecticut Light petitioned for clarification of whether the order was to have prospective effect only. FERC denied rehearing but in clarification stipulated that the rate was to be prospective only. The Connecticut Light & Power Co., Opinion No. 114-A (April 20, 1981), J.A. 872. The result was that the partial requirements customers remain subject to the stratified rate and 100% demand ratchet; the full requirements customers are not subject to the demand ratchet after February 19, 1981, but do not receive refunds for amounts they paid under the ratchet from the date R-4 became effective (February 1, 1979) until February 19, 1981.

II. THE RATE FOR PARTIAL REQUIREMENTS CUSTOMERS

The central question posed in this appeal is simple: whether the rates proposed by Connecticut Light in R-4 for the partial requirements customers are just and reasonable. 16 U.S.C. § 824d(a) (1976). The burden of proof is on the utility to show that they are; during the administrative proceedings, Connecticut Light attempted to meet that burden by showing that the rates charged allocated costs of service among customers in a reasonable manner. See Cities of Batavia, Naperville, Rock Falls, Winnetka, Geneva, Rochelle and St. Charles, Illinois v. FERC, 672 F.2d 64 (D.C.Cir.1982). The ALJ, and FERC, determined that the Company had met its burden with respect to the partial requirements customers. In this appeal, we must determine whether substantial record evidence supports their conclusions. 16 U.S.C. § 825l (b) (1976). Before turning to this task, the central question in this litigation, we must first address several preliminary issues generated by the extensive history of Connecticut Light's efforts to develop a stratified rate.

Although R-4 is not identical to the earlier stratified rate proposals, several problems in the earlier designs recur as issues in this proceeding. CMEGA maintains that these issues were settled in R-1 and R-3, and that collateral estoppel bars their reconsideration here. For example, a premise of assigning the higher charges to peak or off-peak usage is that costs of service can be traced to particular generating facilities with different cost features. Connecticut Light, however, normally allocates the costs of capital facilities system-wide. In R-1, the Commission opined that the Company had "not come to grips with" the problem of tracing costs of service to particular facilities. CMEGA contends that the parties are estopped from challenging this determination in the R-4 proceeding.9

The doctrine of collateral estoppel (issue preclusion) may be invoked in administrative proceedings, with its usual purpose of preventing relitigation of issues earlier aired by the parties and determined by an adjudicatory tribunal. United States v. Utah Construction & Mining Co., 384 U.S. 394, 422, 86 S.Ct. 1545, 1560, 16 L.Ed.2d 642 (1966). See generally Restatement (Second) of Judgments § 83 (1982). The doctrine, however, applies only to relitigation of the same issue, Montana v. United States, 440 U.S. 147, 155, 99 S.Ct. 970, 974, 59 L.Ed.2d 210 (1979). It does not apply when a judgment of policy is reconsidered by an agency in quasi-legislative proceedings. FTC v. Texaco, Inc., 555 F.2d 862, 932-33 (D.C.Cir.), cert. denied, 431 U.S. 974, 97 S.Ct. 2940, 53 L.Ed.2d 1072 (1977). Ratemaking proceedings are especially unlikely to present the proper occasion for invocation of the doctrine, because the appropriateness of a given rate involves policy considerations such as the encouragement of conservation that may be weighed differently over time. Borough of Lansdale, Pa. v. FPC, 494 F.2d 1104, 1115 n.45 (D.C.Cir.1974); see also Florida Power & Light Co. v. FERC, 617 F.2d 809, 816 (D.C.Cir.1980) (agency may adopt new policy in adjudicative proceeding). Economic circumstances, too, may change, and with them the reasonableness of a given rate, Pacific Gas Transmission Co. v. FPC, 536 F.2d 393 (D.C.Cir.), cert. denied, 429 U.S. 999, 97 S.Ct. 527, 50 L.Ed.2d 610 (1976). In addition, the central issue in a rate proceeding is frequently whether the utility has presented evidence sufficient to show that the proposed rate is reasonable. A determination that the utility has not met its burden with respect to one rate does not preclude the utility from making a more successful showing when it files a new rate, as it has the statutory right to do, 16 U.S.C. § 824d(d) (Supp. IV 1980); FTC v. Raladam Co., 316 U.S. 149, 62 S.Ct. 966, 86 L.Ed. 1336 (1942).

In this case, the ALJ determined that the issues singled out by CMEGA in R-4 were not the precise issues that had been resolved in the earlier proceedings, and therefore refused to give the prior determinations collateral estoppel effect. He noted particularly that R-4 was a different tariff, that the R-1 and R-3 tariffs had been rejected by the Commission for failure of proof, and that recent changes in energy policy might bear on Commission reconsideration of stratified rate designs. J.A. 649-52. We have reviewed the issues to which CMEGA seeks to apply collateral estoppel, and we agree with the ALJ that these were not appropriate occasions for invocation of the doctrine. For example, in R-1 the Commission found only that Connecticut Light had not solved the problem of tracing costs of service to particular generating facilities; in R-4, the Company presented a method of making the cost assignments, the success of which must be evaluated in this proceeding. We therefore affirm the refusal to apply collateral estoppel to any of the issues in this case.

CMEGA argues that a number of evidentiary rulings made by the ALJ during the R-4 proceedings deprived it of an adequate opportunity to present its case. We disagree; the ALJ's rulings were reasonable and any difficulties encountered by CMEGA were entirely of its own making.

First, CMEGA objects to the treatment of testimony from the R-3 proceedings. CMEGA sought to have two chunks of material from R-3 introduced as items by reference in R-4; one was admitted, but the second, Item by Reference J, was not, and CMEGA contests its rejection. Item by Reference J consisted of some 650 pages of testimony from R-3 concerning the stratified rate and ratchet design proposed in R-3. J.A. 263. On rebuttal, CMEGA proffered this material as an undifferentiated mass, without prior notice and with the justification that it was relevant because references had been made to R-3 throughout the R-4 proceedings. Id. Both the Company and Commission staff objected, because of the short notice, because the material included testimony from witnesses not even called in R-4, and because CMEGA had failed to make a showing of the specific relevance of the material to R-4. J.A. 264. The ALJ refused to admit Item by Reference J, on the ground that it consisted of a mass of material mostly directed towards the specifics of the R-3 rate and therefore irrelevant in R-4. J.A. 269. He also noted that CMEGA could have sought to introduce the material much earlier, as part of its case in chief, when the nature of the Company's rate proposal was clear. J.A. 270.

The ALJ has the power to make reasonable determinations as to the admissibility of materials in proceedings before him, 5 U.S.C. § 556(d) (1976); cf. Seacoast Anti-Pollution League v. Costle, 572 F.2d 872 (1st Cir.), cert. denied, 439 U.S. 824, 99 S.Ct. 94, 58 L.Ed.2d 117 (1978) (agency may make reasonable determinations as to the need for cross-examination to develop full factual record in adjudicative proceedings). We find the ALJ's decision to exclude Item by Reference J was reasonable and did not deprive CMEGA of a fair opportunity to develop its position.

The second evidentiary ruling of which CMEGA complains concerned Exhibit 26, a set of work papers submitted by the Company in support of the capacity and energy costs it calculated for peak and off-peak production plant. J.A. 36, 44. Exhibit 26 was submitted in the Company's rebuttal testimony, in response to staff comments that they lacked information necessary to evaluate the cost figures. J.A. 151-52. CMEGA objected to the introduction of Exhibit 26 and contended that they had not been given an adequate opportunity to respond to it. J.A. 146. In defense of Exhibit 26, the Company noted that it had been submitted as soon as the staff had made its criticisms known, and that CMEGA had had over two months to prepare a reply. J.A. 151-52. The ALJ denied CMEGA's motion, because he found that Exhibit 26 was an appropriate part of the Company's rebuttal case. This, too, we find to have been a reasonable exercise of discretion by the ALJ.

The final evidentiary decision of which CMEGA complains concerned a study correlating the load shapes of the wholesale customers with the load shape of Northeast Utilities. Connecticut Light is a part of the Northeast system, and its case for the stratified rate depended on showing that the costs of serving the wholesale customers at peak and off-peak periods corresponded to the costs of running the Northeast system at these times. The Commission staff originally opposed the stratified rate, among other reasons because they argued that Connecticut Light had not made an adequate showing of the similarity between the load shapes of the wholesale customers and the Northeast system. In rebuttal, Connecticut Light submitted the correlation study, which was eventually important to the Commission staff's support of the rate and the ALJ's decision adopting it. J.A. 662-63. CMEGA objected to the inclusion of the correlation analysis in the rebuttal testimony, and sought a further opportunity to respond to it. It put forth no indication of what it would do to refute the correlation analysis, however, until it filed comments in response to a settlement offer by Connecticut Light. J.A. 376-77. Recognizing that the matter of the correlation analysis might require further exploration, and determining that such exploration would be accomplished most fairly on the administrative record, the ALJ offered to reopen the record to permit any material the parties wished to submit on the analysis. CMEGA "declined" the invitation to reopen the record, both because it did not want to let the Company submit further materials in defense of the analysis, and because it wanted to use the opportunity to reopen the record to comment on other issues. J.A. 420-22. The ALJ, as a result, did not reopen the record.

We think that the ALJ's actions here were entirely reasonable. The correlation analysis was properly a part of the Company's reply to staff criticism. In spite of CMEGA's delay, the ALJ took extra steps to allow CMEGA to respond to the analysis in a procedurally regularized manner. The ALJ was not obligated to reopen the record for all purposes, or to reopen the record for one side without reopening it for the other side. CMEGA chose not to take advantage of the opportunity it was afforded, and must live with the record as it stands.

CMEGA's final procedural complaint is that the ALJ made several errors in allocating the burden of proof. First, CMEGA contends that the ALJ placed the burden of proof on CMEGA to show that the R-4 rate was unreasonable-rather than placing the burden on the Company to defend the rate, as the statute requires. Not content with ensuring that the Company bear an ordinary burden of proof, CMEGA further contends that in light of the failure of earlier stratified rate proposals, Connecticut Light must bear an especially heavy burden to support its stratification proposal in R-4. We reject both of these contentions.

As to the first contention, CMEGA's allegation that the ALJ placed the burden of proof on CMEGA misdescribes the procedures followed. Under the Federal Power Act, 16 U.S.C. § 824d(e) (1976), the utility is required to bear the burden of showing that a newly filed rate is reasonable. The ALJ found that the utility had presented substantial evidence in support of the rate, and noted that the CMEGA municipalities had failed to present evidence in support of their criticisms. E.g., J.A. 661. This comment by the ALJ was a reply to the municipalities' submissions-most of which relied on their submissions in R-3-but did not shift the burden of proof to CMEGA. The ALJ did state that he was not ruling on the application of the demand ratchet to the partial requirements customers, because "It ha(d) not been raised as an issue by any party," J.A. 672, although he also noted that the staff had agreed that the ratchet was appropriate for these customers because of their ability to reduce their peak demands. J.A. 668. The Commission, however, did not rely on the ALJ's reasoning, but concluded that the ratchet should be applied to partial requirements customers because of their ability to plan to reduce their peak demands. J.A. 835. The Commission also noted that CMEGA had not criticized the Company for applying the ratchet to the partial requirements customers; this was again a reply to CMEGA's case, not a shift of the burden to CMEGA.10

CMEGA also contends that because of the Commission's earlier decisions rejecting stratified rates submitted by Connecticut Light, it must give "more careful consideration" to the Company's submission in R-4,11 CMEGA Brief at 21. CMEGA cites no authority for what appears to be the novel proposition that the Company must make out a stronger case in favor of a rate design once other, similarly structured rates have been rejected, and we reject the proposition.12 Connecticut Light has the right to file a new rate, and to have the rate considered as the statutory scheme provides. Certainly, FERC's earlier decisions concerning rate designs submitted by Connecticut Light or other utilities are relevant as precedent in R-4; where factual issues are the same, evidentiary materials from the earlier proceedings may also be relevant to the later ones, see supra at p. 63. But in R-4, as always, the Commission's duty is to determine whether the Company has met its burden of showing that the rate it proposes is just and reasonable, 16 U.S.C. § 824d(a), (e) (1976), and our duty on review is to determine whether substantial evidence supported the Commission's determinations, id. § 825l (b).

B. Substantive Issues Concerning the Rate Design

CMEGA contends that FERC's treatment of the proposed demand ratchet for the partial requirements customers was an unjustified departure from prior Commission precedent. While the Commission may change policies, it may not do so without reasoned explanation. E.g., Hatch v. FERC, 654 F.2d 825, 834 (D.C.Cir.1981) (agency may alter past policies in adjudicatory context if it provides "reasoned explanation" for the departures). In this case, we find that FERC's treatment of the demand ratchet was consistent with evolving FERC policy on the ratchet issue.

FERC has identified several difficulties in the design of an appropriate demand ratchet. In the first place, ratchets may discourage conservation; if a customer knows he will be billed at a ratcheted amount, he will have little incentive to curb use during periods of lower demand. E.g., Carolina Power & Light Co., Opinion No. 19, 15 Fed. Power Serv. 5-619 (1978). This disadvantage can be overcome, however, if it is desirable to encourage a customer to reduce levels of peak usage, or if it is unlikely that usage patterns are such that off-peak use will expand in response to the ratchet.

Secondly, when combined with the 12-CP method of allocating demand costs among customer classes, a billing ratchet may not track actual costs of service. For example, suppose that some members of the customer class have high noncoincident peak demand which is picked up by the ratchet. Then, members of the customer class will pay a lower demand charge per unit than if the demand charge were unratcheted, because the total demand charge will be divided up among a larger number of units. Customers whose demand occurs at peak will pay a lower total demand charge than they would without the ratchet, whereas customers with high noncoincident peaks will pay a larger overall charge because of their total usage amounts. Customers with high noncoincident peaks will, therefore, be subsidizing customers whose usage coincides with the system peak. See Central Illinois Light Co., Opinion No. 81, 10 FERC P 61,248 (1980) (CILCO ), reversed and remanded on other grounds, Villages of Chatham and Riverton, Illinois v. FERC, 662 F.2d 23 (D.C.Cir.1981). This difficulty, too, can be obviated by characteristics of customer demand; for example, if customers' usage levels tend to peak with the system peak, no such subsidy will materialize.

Recognizing these difficulties, FERC has recently rejected a number of proposed ratchets in combination with the 12-CP method of allocating demand costs. See, e.g., Minnesota Power & Light Co., Opinion No. 86, 11 FERC P 61,312 (1980); CILCO, 10 FERC P 61,248 (1980); Indiana & Michigan Electric Co., Opinion No. 79, 10 FERC P 61,238 (1980). FERC has, however, permitted the combination where a utility can show that "the ensuing disadvantages to consumers of an additionally imposed demand ratchet are outweighed by any benefits to be derived by the utility itself, or by the consuming public." CILCO, 10 FERC P 61,248 (1980); see also Kansas Gas & Electric Co., Opinion No. 80-B, 17 FERC P 61,180 (1981); Union Electric Co., Opinion No. 94, 12 FERC P 61,239 (1980). The municipalities seek to elevate this rule to a presumption against the combination of a ratchet with the 12-CP method of allocation, but it is not. Throughout, FERC has emphasized that the acceptability of a proposed ratchet must rest with the facts of the particular case. In considering whether the ratchet would prove beneficial to Connecticut Light, therefore, FERC did not depart from its evolving precedent with respect to ratchets.13 There remains, however, the question of whether substantial evidence supported FERC's determination that the application of the demand ratchet to the partial requirements customers would prove beneficial in this case.

2. Evidence in Support of the Demand Ratchet

FERC approved the demand ratchet for the partial requirements customers because

On the basis of the record in which both CL&P (Connecticut Light) and the staff have demonstrated some need for the 100% demand ratchet for cost recovery, we shall permit the Company to include this provision in its partial requirements rate.

Opinion No. 114, J.A. at 835-36. The Commission in addition noted that both staff and the Company had relied on the ability of partial requirements customers to utilize their own generating facilities and reduce their amounts of peak demand. Id. at 835.

Evidence in the record, although somewhat sketchy, supports this contention. The Company submitted evidence to show that load shapes of the wholesale customers paralleled those of the system as a whole, which bolsters the contention that it will be beneficial to the Company to encourage the customers to decrease their amounts of peak demand. J.A. 77-78 (Rebuttal Testimony of Richard Brown). A Company witness testified that the ratchet was needed to encourage the customer to define how much of its demand will occur at times of peak usage. J.A. 176 (Testimony of Richard Brown).14 A staff witness also testified that the ratchet was warranted for the partial requirements customers, who can plan to reduce their demands at the system peak, unlike the full requirements customers for whom the ratchet was disallowed. J.A. 276-77 (Testimony of James Gilliam).15 The possibility that the partial requirements customers might utilize alternative power sources was not a mere hypothetical; the Company described in detail the variety of sources available in New England, where power is in oversupply. J.A. 86-88. CMEGA did not reply to this testimony. Finally, although CMEGA suggests the hypothetical of a partial requirements customer with a high noncoincident peak, J.A. 181, it gives no examples of customers who might be expected to bear more than their fair share of demand costs because of the operation of the ratchet. Indeed, the similar climate of the area served by the Northeast system suggests that all of the wholesale customers will have load shapes parallel to the system as a whole, J.A. 199; and record evidence suggests that the ratchet will not substantially redistribute among customers the burden of meeting Connecticut Light's demand costs. J.A. 70.

We conclude, therefore, that there is sufficient record support for the determination that the ratchet would prove useful in encouraging reductions in demand at the time of the system peak, when the burden on the system capacity is the greatest. Moreover, it also appears from the record that the ratchet will not redistribute costs unfairly.16 FERC's decision to approve the ratchet for the partial requirements customers was reasonable, and we affirm.

3. Evidence in Support of the Stratified Rate

Connecticut Light's previous efforts to develop a rate with different demand and energy charges for peak and off-peak usage were rejected. The rejections rested solely on the Commission's determination that the Company had not made an adequate showing that the stratifications tracked actual costs of service. CMEGA contends that the Company's evidence once again falls short, but we find substantial record evidence in support of the Commission's decision that this time Connecticut Light's case for the stratified rate passed muster.

The ALJ found that in R-4, Connecticut Light had succeeded in overcoming the chief difficulties in its earlier efforts to show that the stratified rate accurately reflected actual costs of service, see supra note 7. First, the Company had submitted more extensive evidence backing up the correlation between the load of the Northeast system and the loads of the wholesale customers: a comparison of loads on the day of peak usage in summer and winter, a similar comparison for the week of peak usage in summer and winter, and a statistical analysis of hour-by-hour load shapes in 1978. J.A. 656. Second, the Company had provided further evidence in support of the point at which it had drawn the dividing line between service to be classified as "peak" and "off-peak." J.A. 663. The record as it stands supports the ALJ on these points, and we, like FERC, affirm.

III. REFUNDS FOR THE FULL REQUIREMENTS CUSTOMERS

In the original form submitted by Connecticut Light, R-4 included a demand ratchet for both the full and the partial requirements customers. The ALJ's initial decision disallowed the ratchet for both full and partial requirements customers and ordered refunds to be paid to both. J.A. 672. FERC, in reversing the ALJ on the application of the ratchet to the partial requirements customers, specified a method for dividing up demand costs between full and partial requirements customers, in order to ensure that only the latter were subjected to the ratchet. It did not, however, specify whether the new method of calculating demand costs for the wholesale customers was to be applied prospectively only. Opinion No. 114, J.A. 836. After Connecticut Light petitioned for clarification, FERC on rehearing decided that the new rate design should be applied prospectively only, to protect the Company against the risk of undercollection. Opinion No. 114-A, J.A. 874.17 The municipalities contend that this decision was error and that the full requirements customers should be given refunds for the approximately two years that the Company had collected under the ratchet. We disagree.

Under the Federal Power Act, 16 U.S.C. § 824d(e) (1976), the Commission may order refunds of portions of rates found unjustified. Refunds are not mandatory; the Commission has discretion to decide whether a refund is warranted in light of the interests of the customer and the utility. See Belco Petroleum Corp. v. FERC, 589 F.2d 680, 686 (D.C.Cir.1978). The Commission advanced two reasons here for making the rate prospective only: that the Company might be subject to undercollections from the refund because it could not collect retroactively from other customers, and that retroactive changes in rates cannot affect customer demand. Opinion No. 114-A, J.A. 874. These reasons were offered by the Commission in an earlier case in which it refused a refund to customers that had paid under a ratchet, and this court affirmed. Commonwealth Edison Co., 19 Fed. Power Serv. 5-75 (1979), aff'd, Cities of Batavia, Naperville, Rock Falls, Winnetka, Geneva, Rochelle and St. Charles, Illinois v. FERC, 672 F.2d 64, 85 (D.C.Cir.1982).

As in Cities of Batavia, id., we find that the Commission exercised its discretion reasonably in deciding to apply the approved rate prospectively. Earlier stratified rate designs submitted by Connecticut Light had been disallowed by FERC, but only because of insufficient evidentiary support. Many other issues in R-4 had been settled, and the Commission was properly concerned to see that Connecticut Light did not suffer additional revenue shortfalls beyond those in the settlement agreement. J.A. 873. We therefore find that the Commission gave adequate attention to the interests of both the Company and CMEGA, and we affirm.

This is a case in which a utility company, faced with declining demand, initiated an innovative rate design. FERC found that the utility, after several efforts, had finally submitted a version of the design that demonstrably tracked costs of service in an accurate manner. After careful review, we agree, and the decision of the Commission is therefore

Affirmed.