Cordy, J.

On December 15, 2003, the Commissioner of Insurance (commissioner) released the opinion, findings, and decision on 2004 private passenger automobile insurance rates, (2004 decision) establishing Massachusetts automobile insurance rates for 2004. The Attorney General sought judicial review in the Supreme Judicial Court for Suffolk County of several aspects of the 2004 decision and a related discovery order. A single justice reserved and reported the case to the full court without decision. The Attorney General challenges (1) the adoption of a new methodology for modeling underwriting profits, (2) the calculation of premium charges for policyholders who elect increased bodily injury limits, and (3) the commissioner’s denial of his motion to compel discovery of certain data related to insurance agency expenses.

We conclude that the commissioner’s choice of an underwriting profits model that uses insurers’ internal rates of return (IRR) is adequately explained and reasonably supported by evidence in the administrative record, and the Attorney General’s motion to compel discovery of agency expenses is moot. We further conclude that the commissioner’s rejection of the Attorney General’s evidence that premiums charged policyholders with increased bodily injury limits are excessive lacks a reasoned basis. We remand the portion of the 2004 decision that establishes premium charges for purchasers of optional bodily injury coverage to the commissioner for reconsideration, further findings, and, if necessary, further proceedings.

1. Background. The commissioner may set automobile insurance rates (also termed “premium charges”) on an annual basis if she finds, after investigation and public hearing, that unregulated competition in the Massachusetts market for automobile insurance may lead to excessive rates or might be destructive of competition or insurers’ solvency. G. L. c. 175, *795§ 113B. G. L. c. 175E, § 5. Massachusetts Auto. Rating & Acc. Prevention Bur. v. Commissioner of Ins., 401 Mass. 282, 284 (1987) (Mass. Auto. Rating I). After a public hearing on May 16, 2003, the commissioner determined that market conditions required rate setting for private passenger automobile insurance for 2004. Consequently, the commissioner initiated a rate setting process for the purpose of generating “adequate, just, reasonable and nondiscriminatory premium charges,” as is her obligation under G. L. c. 175, § 113B. That process involved the examination of four separate aspects of the rates — (1) the safe driver insurance plan; (2) the underwriting profits provision; (3) cost containment and fraudulent claims; and (4) the main rate — each with its own proceeding at the Division of Insurance. Parties to these four proceedings included the Automobile Insurers Bureau of Massachusetts (AIB), the State Ratings Bureau (SRB), the Massachusetts Association of Insurance Agents (MALA), and the Attorney General.2 They submitted filings reflecting their respective recommendations and presented the testimony of supporting witnesses.

The 2004 decision details the commissioner’s determinations on each of the four litigated aspects of the rates, only two of which — the underwriting profits provision and the main rate — are relevant to this appeal.

2. Standard of review. In reviewing the commissioner’s determinations in rate setting proceedings, this court examines whether they have reasonable support in the evidence. Automobile Insurers Bur. of Mass. v. Commissioner of Ins., 415 Mass. 455, 457 (1993); Aetna Cas. & Sur. Co. v. Commissioner of Ins., 408 Mass. 363, 378 (1990) (Aetna Cas. & Sur. Co.); Mas*796sachusetts Auto. Rating & Acc. Prevention Bur. v. Commissioner of Ins., 384 Mass. 333, 337 (1981) (Mass. Auto. Rating II). This standard parallels the substantial evidence standard, requiring us to be satisfied that a “reasonable mind might accept [the evidence] as adequate to support a conclusion.” G. L. c. 30A, § 1 (6). See Aetna Cas. & Sur. Co., supra at 378 n.14.

We do not substitute our judgment for the commissioner’s where substantial evidence in the record supports her conclusions. Automobile Insurers Bur. v. Commissioner of Ins., supra at 457-458; Mass. Auto Rating II, supra at 342. We recognize that, to make final determinations in the rate setting context, the commissioner has wide discretion in weighing conflicting evidence and gouging the credibility of witnesses. Massachusetts Ass’n of Ins. Agents v. Commissioner of Ins., 425 Mass. 477, 484 (1997). Aetna Cas. & Sur. Co., supra at 375. We also give deference and “ ‘due weight to the commissioner’s experience, technical competence, specialized knowledge, and discretionary authority’ . . . particularly appropriate when reviewing her choice of methodology.” Automobile Insurers Bur. of Mass. v. Commissioner of Ins., 430 Mass. 285, 296 (1999), quoting Aetna Cas. & Sur. Co., supra at 378-379.

At the same time, the commissioner must make “findings [that] indicate the over-all basis of [the] decision.” Aetna Cas. & Sur. Co., supra at 374. See Mass. Auto. Rating I, supra at 287. “Adequate explanation from the commissioner will ensure that our review is performed on the basis of what we know rather than what we might opine from a perusal of the record.” Id. at 288. “[T]he commissioner need not make findings on every controverted issue of fact or law so long as [her] findings indicate the over-all basis for [her] decision and permit effective appellate review.” Massachusetts Ass’n of Ins. Agents v. Commissioner of Ins., supra at 484, quoting Aetna Cas. & Sur. Co., supra at 374.

3. Underwriting profits provision. The underwriting profits provision of the rates3 takes into account the money earned by insurers between the time premiums are collected and the time claims are paid. During this period, insurers invest the premiums *797they have received from policyholders and earn a return on those investments. Because, in general, “money is made on investments, not on underwriting,” Massachusetts Auto. Rating & Acc. Prevention Bur. v. Commissioner of Ins., 381 Mass. 592, 605 (1980), the underwriting profits provision (often referred to as “allowance”) adjusts automobile insurance rates to recognize the role of investment income in the business of insurance underwriting.4 To determine an underwriting profits allowance that balances the insurers’ interest in earning a fair profit on their businesses and the interest of Massachusetts policyholders in reasonable rates, the commissioner must employ a mathematical model that analyzes financial risks and predicts insurer cash flows throughout the coming year, using projections based on the historical experience of the property casualty industry.

a. Commissioner’s choice of models. From 1990 to 2003, the commissioner employed the Myers-Cohn model, a methodology focused on the flow of premium charges paid by policyholders and payments by insurers to cover policyholder losses. In the 2004 decision, the commissioner utilized, for the first time in Massachusetts automobile insurance rate setting, a model that used the insurers’ “internal rate of return” (IRR model) for calculating an appropriate allowance for underwriting profits. The key input to the ERR model is the cost of capital, a measure of the rate of return companies could expect to earn by investing in another business of equivalent risk to insurance underwriting. By contrast, the Myers-Cohn model uses the “beta of liabilities,” a measure of how policyholders’ return on their premium charges, i.e. loss, expense, and tax payments made by insurers on their behalf, compares to a market portfolio of other risky assets. In other words, the Myers-Cohn model and the *798IRR model calculate the underwriting profits allowance differently by observing and then predicting the cash flow of the automobile insurance industry from different vantage points — Myers-Cohn from the perspective of the policyholder and IRR from the perspective of insurers’ shareholders. See Cummins, Multi-Period Discounted Cash Flow Ratemaking Models in Property-Liability Insurance, 57 J. Risk & Ins. 79 (1990) (describing differences between IRR model and Myers-Cohn model). Both the AIB and the SRB proposed using differing IRR models for calculating the underwriting profits allowance, while the Attorney General urged retention of the Myers-Cohn model.

Once a model is selected, the outcome further depends on the values placed on its various inputs. Although the commissioner agreed to adopt an IRR model, she rejected several input values proposed by the AIB and SRB, arriving at a final underwriting profits allowance of -.69% a figure falling between the extremes of AIB’s recommendation of +3.38% and the Attorney General’s recommendation of -2.91%.

The Attorney General contends that the commissioner’s 2004 decision not only fails to contain adequate findings supporting the choice of an IRR model, it actually contains findings discrediting that choice. We disagree with the Attorney General’s interpretation. The 2004 decision adopts a composite IRR model assembled from methodologies proposed by the AIB and SRB and from input values proposed by all parties, including the Attorney General and the commissioner. This combination of methodologies and inputs, and the ultimately determined underwriting profits allowance, are distinct from the AIB and SRB’s IRR proposals. A careful reading of the 2004 decision makes apparent that the criticisms it contains are of the AIB and SRB IRR models, and not of the ultimate methodology selected by the commissioner.

Relying on testimony from experts offered by the AIB, the SRB, and even the Attorney General, the commissioner found that “a properly implemented IRR model” could yield reasonable rates meeting the statutory standard of G. L. c. 175, *799§ 113B.5 Her extended analysis and adjudication of the disputed input values contain further findings supporting her choice of methodology. In these circumstances, the commissioner has provided an “[ajdequate explanation” for the 2004 decision’s underwriting profits allowance. Mass. Auto. Rating I, supra at 288.

In addition, the commissioner’s determination to use the IRR model finds reasonable support in the voluminous record of the case. Admittedly, that record contains competing testimony by actuarial experts and disputed data about the relative merits of the Myers-Cohn model and the IRR model. The Attorney General points to evidence that the commissioner’s model choice will allegedly yield an $80 million windfall for insurers; the AIB counters that the evidence suggests the Attorney General is overestimating investment returns in an age of depressed interest rates. On one side of the issue, Professor Richard A. Cohn, the codeveloper of the Myers-Cohn model, testified that it “is and always has been a fair premium model, designed to produce a fair premium for both policyholders and shareholders. . . . [I]t is superior to the AIB IRR model, which views profit and return entirely from the shareholder perspective.” On the other side, Dr. Richard A. Derrig testified that the Myers-Cohn model “calculates the present value of policyholder costs and assumes there is an operating market that can charge these prices. The IRR model . . . is a much more appropriate way for the [cjommissioner to monitor the companies’ prices than [Myers-Cohn].” It is the commissioner, however, not the court, that is statutorily charged with resolving these types of disputes. As we have noted, her decision to reject or adopt an underwriting profit model is owed substantial deference:

*800“In weighing the relative merits of different methods of calculating the underwriting profit allowance, the Commissioner has considerable latitude. There is no legal or constitutional requirement that a particular theory or method be used in determining the profit allowance to be used in computing rates. . . . [W]e decline the bureau’s clearly implied invitation to compare both models in light of all the evidence and to select the one we decide is better. To accept this invitation would amount to usurpation of the Commissioner’s authority.” (Citation omitted.)

Mass. Auto. Rating II, supra at 341-342. See Massachusetts Auto. Rating & Acc. Prevention Bur. v. Commissioner of Ins., 389 Mass. 824, 831-833, 837 (1983) (affirming commissioner’s choices of hypothetical insurance company and surplus allocation models for setting underwriting profits allowance). On the record of this proceeding, our deference “to the Commissioner’s experience and expertise in [choosing her] methodology,” id. at 837, is fully warranted.

b. Reasoned consistency. That the commissioner is replacing a long-standing methodology with a provisional, “untested” one does not embolden us to require the commissioner to retain the Myers-Cohn model.6 Because the commissioner’s decision had its basis in substantial evidence, we reject the Attorney General’s argument that the 2004 decision’s choice of an IRR model is a departure from the “reasoned consistency” this court requires of administrative decision-mating. Mass. Auto. Rating I, supra at 287, quoting Boston Gas Co. v. Department of Pub. Utils., 367 Mass. 92, 104 (1975). “Reasoned consistency” cannot be held to paralyze an agency that, within its statutory discretion and on substantial evidence, decides to respond to new conditions by replacing dated methodologies. See MCI WorldCom Communications, Inc. v. Department of Telecommunications & Energy, ante 103, 116 (2004) (“[reasoned consistency] does not mean that the department may never deviate from its original position, only that any such change must be explained”). Here, the commissioner explains her choice of an IRR model as a change in policy directed toward *801improving a market climate where the number of insurance companies willing to underwrite automobile policies in Massachusetts is declining.

The commissioner has, over time, implemented the “reasoned consistency” obligation by requiring proposed methods to be superior to those they replace. The Attorney General argues that the commissioner has failed to demonstrate the superiority of the IRR model in the 2004 decision and urges us to overturn its choice on that basis. While the commissioner does not make an explicit finding that the adopted IRR model will be superior,7 the 2004 decision amply implies that conclusion with its critique of the Myers-Cohn model, its openness to alternatives,8 and its ultimate adoption of an IRR model.9 Such a conclusion finds substantial support in the evidence. See Aetna Cas. & Sur. Co., *802supra at 375 (“the record . . . contains testimony which independently provides reasonable evidentiary support for [the] decision”).

Unlike the Myers-Cohn model, which was developed specifically for rate setting, the IRR model enjoys compatibility with accessible concepts of corporate finance. For example, the “cost of capital” measure so prominent in IRR models is ubiquitous in the financial marketplace, while Myers-Cohn’s key input — the “beta of liabilities” — is not. This is particularly significant in the context of the commissioner’s discussion of the declining number of insurers willing to underwrite automobile insurance policies in Massachusetts and the need for rate setting reforms. The adoption of an IRR model marks a step, albeit an initial and provisional one, toward the commissioner’s goal of encouraging new insurers to enter the Massachusetts automobile insurance market.10 For these reasons, the commissioner’s decision to adopt an IRR model for calculating the underwriting profits provision of automobile insurance rates satisfies the “reasoned consistency” standard for administrative decision-making.

4. Bodily injury increased limits factors. Within the context of the main rate proceeding, the commissioner uses “increased limits factors” (ILFs) to set the premium charges that insurers may charge policyholders who elect to buy coverage that *803exceeds the statutory minimum. Such additional coverage provides for increased limits on the payment of claims for property damage, uninsured or underinsured drivers of covered vehicles, and bodily injury (BI), the only coverage at issue here. The statutory mínimums for BI coverage (basic limits) are $20,000 per person, $40,000 per accident. To determine proper premium charges for policyholders with additional coverage (e.g., $100,000 per person and $300,000 per accident, or 100/300 policies), the commissioner calculates ILFs as the ratios between the total annual losses incurred by each group of policyholders (e.g., $221 million by 100/300 policyholders in 2002) and the annual losses incurred by those same policyholders within the basic limits (e.g., $149 million by 100/300 policyholders in 2002). Calculated with historical and projected data, these ILFs (e.g., 1.4625 for 100/300 policies in the 2004 decision) are then multiplied by the basic premium charges that cover losses incurred by all policyholders within the basic limits to establish the premium charges for policyholders who buy additional BI coverage.11 ILFs are intended to ensure that premium *804charges for increased coverage compensate insurers for the expected losses of the pohcyholders who purchase that coverage.

In the 2003 rate setting proceeding, the Attorney General challenged the prevailing methodology for calculating BI ILFs. The Attorney General’s expert, Allan I. Schwartz, testified that the formula improperly calculates ILFs based on the loss history of one subset of pohcyholders (those at the given level of increased limits) and applies those ILFs to an average basic premium derived from the loss experience of all pohcyholders (ILF apphcation mismatch). Because increased limits pohcyholders in general incur fewer losses than those pohcyholders who purchase basic coverage, the Attorney General argued that using the over-all average basic premium in the formula overestimates the cost of increased limits pohcyholders’ losses. This error contributes to an unmonitored subsidy that, unlike cross-subsidies between pohcyholders on a basic Emits basis,12 insurers retain, making the rates for increased Emits coverage excessive (subsidy imbalance). The commissioner rejected the Attorney General’s 2003 proposals to adjust the formula to reduce the unintentional subsidy, if any, finding that the Attorney General had not met the burden of demonstrating the adjusted method’s superiority.

The Attorney General renewed his objection to the calculation of BI ILFs in the 2004 rate setting proceeding, and his filing provides testimony and other evidence specifically address*805ing the concerns raised in the commissioner’s commentary of the previous year.13 It also includes Schwartz’s testimony that *806the overcharges for increased limits policyholders could be reduced if (1) the AIB rebalanced the rates it filed “on [both] a total limits and excess limits basis” by applying a correction factor similar to the balancing adjustment the AIB already performs on a basic limits basis, and (2) the commissioner applied BI ILFs to the appropriate average of increased limits policyholders’ actual losses, rather than the broader average of all policyholders’ loss experience.

In the 2004 decision, the commissioner again rejects the Attorney General’s proposals to correct the subsidy imbalance and ILF application mismatch, repeating the conclusion that the Attorney General has not met the necessary burden of demonstrating the existence of the problem and the superiority of the adjusted methods. She further concludes that the Attorney General did not address the impact of the adjusted methods on the purchasing patterns of consumers, who might flock to lower-priced increased limits coverage.14

*807In this appeal, the Attorney General argues that (1) the commissioner’s conclusion that the evidence did not adequately demonstrate the presence of the subsidy imbalance or the BI ILF application mismatch is not supported by substantial evidence; (2) the commissioner’s concern about an adjustment to the BI rates affecting purchasing patterns similarly has no basis in the record evidence; and (3) the commissioner erroneously failed to adopt his proposed changes to the calculation of increased limits policyholders’ rates, resulting in excessive rates. The commissioner argues that the testimony of the AIB’s expert witness, Kim A. Scott, provides ample support for the commissioner’s decision to continue using the allegedly flawed methodology.

In sharp contrast to the exhaustive vetting of the choice of an IRR model for use in setting the underwriting profits provision of the rates, the explanation of the commissioner’s choice to ignore Schwartz’s testimony and readopt the BI ILF methodology of past years involves little analysis. Instead, the commissioner summarily concludes that the Attorney General has failed to meet the requisite burden. And while the commissioner proceeds to question the impact of the Attorney General’s proposals on consumer purchasing patterns, she does not acknowledge or address the Attorney General’s contrary evidence on that issue.15

We have consistently refused to “supply a reasoned basis for the agency’s action that the agency itself has not given.” Costello v. Department of Pub. Utils., 391 Mass. 527, 536 (1984), quoting Bowman Transp., Inc. v. Arkansas-Best Freight Sys., 419 U.S. 281, 285-286 (1974). In Mass. Auto. Rating I, supra, this court remanded the commissioner’s decisions excluding historical evidence of past underwriting results and adopting a new model to impute Federal tax liability on insurers’ income within the underwriting profits model. In finding the commis*808sioner’s explanations for these decisions lacking, we noted, “[t]he commissioner ‘should not leave counsel and the courts without the guidance of proper findings ... to determine from a voluminous record . . . whether [her] conclusions can be sustained on the evidence.’ ” Id. at 288, quoting Insurance Rating Bd. v. Commissioner of Ins., 359 Mass. 111, 118 (1971). Where we can only speculate as to the rationales for the commissioner’s decision, as with the summary rejection of the Attorney General’s arguments about the subsidy imbalance and the BI ILF application mismatch, “effective review is frustrated.” Mass. Auto. Rating I, supra at 292.

Moreover, the record here does not “independently provided reasonable evidentiary support for [the] decision.” Aetna Cas. & Sun Co., supra at 375. Scott’s equivocal testimony in the rate setting proceeding provides scant support for the commissioner’s summary conclusions. Scott merely testified that, although the subsidy imbalance exists, it is not a “particularly large problem,” and that making adjustments to the rates to avoid overcharging increased limits policyholders would be “extremely difficult to pursue” and may involve unanticipated effects. Set against the Attorney General’s extensive evidence on this issue,16 Scott’s opinions hardly qualify as “substantial evidence” supporting the commissioner’s decision, a decision which is apparently premised on a finding that the subsidy imbalance and BI ILF application mismatch do not even exist.

Although the Attorney General unquestionably “bore the burden of persuasion on the adjustment,” Automobile Insurers Bur. of Mass. v. Commissioner of Ins., 430 Mass. 285, 294 (1999), the commissioner has not adequately explained the “over-all basis” for her decision to reject the Attorney General’s proposed changes to the BI ILF methodology. Aetna Cas. & Sun Co., supra at 374. Mass. Auto. Rating I, supra at 287. Consequently, we are unable to appreciate how or why the commissioner determined that the Attorney General failed to meet that burden. On the record before us, we can neither affirm the commissioner’s rejection of the Attorney General’s concerns nor order the recalculation of rates using the Attorney General’s adjusted methodology. We therefore remand the issue of the al*809leged subsidy imbalance and the BI ILF application mismatch to the commissioner for further findings, and, if necessary, further proceedings. See id. at 293.

5. The commissioner’s discovery order. Finally, the Attorney General asks this court to overrule the commissioner’s denial of a motion to compel discovery of a 2002 MAIA study compiling data regarding the expenses of its member insurance agencies. We conclude that the Attorney General’s appeal from the commissioner’s discovery order is moot because the Attorney General has not challenged the aspect of the 2004 decision to which the requested study pertains.

In the 2003 and 2004 rate setting proceedings, MAIA intervened in the main rate proceeding and submitted advisory filings recommending a dollar amount for the average agent commission expense pure premium (CEPP), another provision of automobile insurance rates that accounts for the reasonable expenses agents incur in selling insurance policies. G. L. c. 175, § 113B. As a part of the 2003 proceedings, MAIA proposed a CEPP on the basis of a 2002 study of agency costs that MAIA conducted with the assistance of consulting actuaries. The commissioner had concerns about the reliability and completeness of the study, ultimately rejecting MAIA’s proposed CEPP, which added a significant upward adjustment to the results of the cost study. The commissioner found the study so substantially and fundamentally flawed that the 2003 decision stated, “the weight that can be given the conclusions MAIA draws by the survey is severely limited.” Despite the study’s flaws, the commissioner “judgmentally” set the CEPP at $113, using the study’s cost estimates for 2001 as “the best recent evidence” on agency costs and applying a small upward adjustment to account for inflation during the intervening years. In the 2004 decision, the commissioner again “judgmentally” set the CEPP at $114, applying another small inflation adjustment to the 2003 amount. In both proceedings, the commissioner found that the evidence did not adequately support the parties’ recommendations, but nonetheless faced the statutorily mandated task of setting a dollar value for the CEPP.

Given the relevance of the cost study to the 2003 CEPP, the Attorney General sought discovery of certain data omitted from *810the cost study from MAIA during the 2004 rate setting proceedings. After MALA indicated that the omitted data was not in its possession, the Attorney General filed a motion to compel its discovery, pursuant to 211 Code Mass. Regs. § 77.06(3) (1997). The commissioner denied the motion in a summary order.

The time for the Attorney General’s questions about the MAIA’s 2002 study has passed, and the narrow appeal from a discovery order does not revive them. The Attorney General did not appeal from the 2003 CEPP that relied on data from the 2002 cost study, nor does the Attorney General appeal from the 2004 CEPP here. The Attorney General’s claim that the whole 2004 decision is technically before us, including the 2004 CEPP, ignores that the complaint in this case asks us to review the commissioner’s discovery order and does not challenge the 2004 CEPP itself on substantial evidence or other grounds. Likewise, the Attorney General’s briefs in this case focus solely on vacating the discovery order and do not argue that we should vacate the 2004 decision’s CEPP as a result of the missing data.

Because the Attorney General did not allege that either CEPP failed to meet the standards of G. L. c. 175, § 113B, the Attorney General has not alleged harm to the parties or to the consumers whom he represents. “Litigation ordinarily is considered moot when the party claiming to be aggrieved ceases to have a personal stake in its outcome.” Acting Superintendent of Bournewood Hosp. v. Baker, 431 Mass. 101, 103 (2000), quoting Attorney Gen. v. Commissioner of Ins., 403 Mass. 370, 380 (1988). Here, the Attorney General lacks a personal stake in the outcome of an appeal from the discovery order.

6. Conclusion. We remand the case to the single justice, who is directed to issue an order vacating so much of the 2004 decision establishing premium charges for policyholders purchasing optional bodily injury coverage and affirming the 2004 decision in all other respects. The single justice is further directed to remand the vacated portion of the 2004 decision to the commissioner for reconsideration in light of this opinion, further findings, and, if necessary, further proceedings. Any party aggrieved *811by the commissioner’s further findings or actions may file a petition for review with the single justice pursuant to G. L. c. 175, § 113B.

So ordered.

Notes

The Automobile Insurers Bureau of Massachusetts (AIB) is an unincorporated voluntary association of the State’s automobile insurers, authorized by statute to represent their interests in rate setting proceedings before the Division of Insurance. G. L. c. 175A, §§ 8 et seq. The State Ratings Bureau (SRB) is itself a unit of the Division of Insurance and participates in rate setting proceedings pursuant to division regulations. 211 Code Mass. Regs. § 77. 02 (1997). The Attorney General intervened in the proceeding as of right, pursuant to G. L. c. 12, § 11F, to represent “consumers in connection with [a] matter involving the rates, charges, premiums and prices charged” by insurers. The Massachusetts Association of Insurance Agents (MAIA) is an association of insurance agents that intervened solely in the main rate proceeding to represent its members’ interest in their commission allowance.

The two other basic provisions of automobile insurance are allowances for payments of claims (losses) and nonclaim administrative expenses. See Mas*797sachusetts Auto. Rating & Acc. Prevention Bur. v. Commissioner of Ins., 401 Mass. 282, 284 (1987) (Mass. Auto. Rating I).

To reflect favorable investment conditions and to avoid excessive premium charges for policyholders, the underwriting profits allowance has in recent years been set at a negative value. This negative value is applied to the rates that would compensate insurers for losses and expenses, resulting in insurers collecting less in premium charges than their anticipated losses and expenses. Both this deficiency and the insurers’ profit margins are made up by the earnings received from invested premiums.

The AIB’s expert, Dr. Richard A. Derrig, testified in this proceeding that there is nothing inherent in model choice that guarantees reasonable underwriting profits allowances for policyholders or insurers. Different models “can produce the same result” given the choice of consistent and reasonable parameters for their implementation. Professor Richard A. Cohn, who testified for the Attorney General on behalf of the model he helped develop, indicated that “internal rate of return models are . . . financially and theoretically sound,” and that the Myers-Cohn model “can be viewed as an internal rate of return model” because both require similar information, such as accurate discount rates for losses and expenses.

While its use may be long standing in Massachusetts, the Myers-Cohn model is not used by any other State in the setting of automobile rates.

The omission of such an explicit finding is somewhat surprising in light of the decision of the commissioner several months earlier approving workers’ compensation insurance rates by similarly departing from the long-standing use of a Myers-Cohn model and adopting an IRR model. Contained in that decision are several explicit findings of the IRR model’s superiority. Decision and Order on the Rate Filing of the Workers’ Compensation Rating and Inspection Bureau No. R2003-08, at 37-41 (Aug. 29, 2003) (“We are persuaded that the ERR model is conceptually simpler than the [Myers-Cohn] model, permits a more direct, and [better] understanding of the relationship between insurers’ financial structures and underwriting profit provisions in rates and therefore, from that point of view, is superior”) (“we find that use of an ERR model in Massachusetts is superior to continued use of the [MyersCohn] model”).

For example, the 2004 decision finds, “[W]e are not persuaded that retention of the unique Myers-Cohn model is necessary to protect Massachusetts consumers. That the Myers-Cohn model has, over time, been extensively addressed in Massachusetts is not sufficient to justify its continued use .... [W]e conclude that it is appropriate to explore the use of an alternative model for the underwriting profits provision that would be superior to the MyersCohn model.”

The 2004 decision hesitates to commit the commissioner to unequivocally endorsing the ERR models’ superiority in the automobile insurance rate setting context. This hesitation appears to flow from a reluctance to approve the model on a “going forward basis” (“That our decision this year is based on application of an IRR model should not be viewed as a commitment to its use in the future”). We conclude that the commissioner’s decision to explore the advantages of implementing an IRR model during the coming year supports her expressed and valid goal of developing a profits model that will “make the ratesetting process simpler and clearer . . . [to] help encourage entry into the market” and be “understandable, principled, and capable of producing results that are fair to policyholders and insurers.”

That the commissioner’s chosen methodology is novel and provisional does not persuade us to disturb the 2004 decision’s choice of an IRR model. See Attorney Gen. v. Commissioner of Ins., 370 Mass. 791, 820 (1976) (“we are mindful that the Commissioner’s approach to the profit allowance is not only novel but complicated, and that somewhat greater imprecision must be tolerated in its initial application than might be acceptable in later years”). The Attorney General alleges that the presence of what the 2004 decision describes as “unresolved issues about the structure of and inputs to an IRR model” required the commissioner to reject it. Here, the Attorney General misreads the 2004 decision, which noted in the quoted passage that a complete treatment of the differences between the AH3 and SRB IRR models was unnecessary in light of the commissioner’s rejection of those IRR models on a “going forward basis.” The 2004 decision proceeds to analyze the methodologies and inputs to its composite ERR model over forty-four pages, sufficiently resolving necessary issues for the coming year. Given the 2004 decision’s forward-looking openness to alternatives, the commissioner did not wish to “quell discussion” of new approaches, particularly those that may be developed by the parties in collaborative workgroups during the course of the year.

For example, premium charges for policyholders who purchase 100/300 coverage for BI are calculated as follows.

Using historical and projected data, the commissioner sets BI “basic pure premiums,” i.e., the average costs of BI losses within the basic limits for all policies, per policy. These are the foundation of the basic premium charges all Massachusetts policyholders pay for their basic BI coverage. As a part of this process, the commissioner adjusts basic pure premiums depending on policyholders’ territories and driving histories, creating public policy cross-subsidies between geographic regions and classes of drivers. See note 12, infra.

The commissioner then assembles two sets of historical data on the value of losses for 100/300 policyholders: (1) claims of 100/300 policyholders that do not exceed the basic limits (basic limits losses) and (2) claims of 100/300 policyholders that exceed the basic levels of coverage (increased limits losses). After adjusting these historical figures to estimate the actual losses insurers can expect to pay in the coming year, the commissioner takes the sum of both loss figures (total losses) and divides that sum by the basic limits losses. The resulting ratio is the BI ILF for 100/300 policies: 1.4625 in the 2004 decision.

The 100/300 ILF is then multiplied by the full set of basic pure premiums, resulting in the pure premiums for 100/300 policies — the part of the ultimate premium charges that is intended to cover the losses of 100/300 policyholders. Pure premiums for 100/300 policies are then combined with other provisions of the rates — the underwriting profits allowance and the allowance for insur*804ers’ nonclaim expenses — to generate the actual premium charges for policyholders who purchase 100/300 coverage for BI.

Pursuant to the commissioner’s statutory mandate to fix and establish automobile insurance rates that are “adequate, just, reasonable and nondiscriminatory,” the long-standing methodology for setting fair premium charges involves the division of policyholders by geographic location into territories and by level of driving experience into rating categories. G. L. c. 175, § 113B. The commissioner then levels some of the rate differences between rating categories and territories and caps spikes in rate increases that would occur if policyholders in each rating category and territory had to cover their own losses. Through this process, automobile insurance regulation results in “public policy” subsidies from drivers with excellent and long driving records and drivers who live in less accident-prone areas to policyholders who have short or accident-prone driving records and policyholders who live in accident-prone areas. See, e.g., Opinion, Findings and Decision on 2003 Private Pas-. senger Automobile Insurance Rates, Nos. R2002-03, R2002-04, R2002-05, at 51-53 (Dec. 23, 2002) (2003 decision).

For example, the Attorney General’s filing offered the following specific responses to the concerns the commissioner had articulated in her 2003 decision.

(1) In response to the commissioner’s suggestion that it was not necessarily unreasonable to apply BI ILFs to an average rate reflecting the losses of all policyholders, Schwartz testified that ILFs should then be calculated using basic limits losses of all policyholders. In other words, the ILFs and the rates to which they are applied should be consistent; if they were consistent, rates for increased limits policyholders would be lower.

(2) In response to the commissioner’s suggestion that the availability of increased limits coverage in all territories and across rating categories served to mitigate the possibility of a subsidy imbalance, the Attorney General’s filing points out that property damage liability ILFs incorporate an adjustment factor (a market mix adjustment factor) that reflects the reality that increased limits customers are disproportionately in lower risk categories and territories. Because it is undisputed that purchasing patterns reflect the same distribution of policyholders in bodily injury coverage, the Attorney General further contends that there should be no barrier to the commissioner’s assuming a similar purchasing phenomenon in order to assess the presence of subsidy imbalances in the context of BI ILFs.

(3) In response to the commissioner’s concern that the Attorney General’s adjustments would endanger valid “public policy” cross-subsidies, see note 12, supra, Schwartz’s testimony isolates the subsidy provision of the rates through algebraic manipulation, showing that it is possible to eliminate the prevailing method’s unintentional subsidy of insurers but preserve cross-subsidies.

(4) In response to the commissioner’s concern that the Attorney General’s proposal would base BI ILFs on “the loss history of only a subset of consumers,” the Attorney General’s filing clarifies that the current methodology for calculating BI ILFs has the same feature, as the factor itself is a ratio of the basic and total loss experience of each purchasing group.

(5) In response to the commissioner’s faulting the Attorney General’s proposal for failing to analyze the possibility of “a substantial shift in purchasing patterns” as a result of the potential reduction in the premium charges for increased coverage, Schwartz extrapolated the impact of the Attorney General’s proposals on previous rate years and testified that the necessary adjustments to the rates would have been stable over time, thereby negating the commissioner’s concern that the adjustments could generate dramatic annual swings in purchasing patterns. The Attorney General’s filing also notes that the commissioner has never before required proponents of methodological changes to analyze the purchasing pattern impacts of those changes and has not put a similar burden on the AIB’s 2004 proposals to increase rates in other areas.

(6) In response to the commissioner’s finding that increased limits policyholders already receive rate reductions for good driving records and for *806living in lower-rated territories as evidence that the current methodology yields reasonable rates, the Attorney General’s filing points out that the subsidy imbalance cuts across rating categories and territories and renders the rates of all increased limits customers too high by varying degrees.

(7) In response to the commissioner’s concern that the Attorney General’s adjustments would not necessarily ensure the accuracy of the ILFs, the Attorney General’s filing contends that no projections in rate setting proceedings can ever provide a guarantee of accuracy, and includes Schwartz’s testimony that, despite this lack of absolute certainty:

“[W]ithout a doubt ... the method proposed by the [Attorney General] can be expected to produce more accurate BI ILFs than the method that is currently being used, and that if the current method continues to be used, then the rates charged to purchasers of excess limits coverage are expected to be excessive, which will result in a subsidy from private passenger automobile insurance consumers in Massachusetts to insurance companies.”

In addition, the Attorney General points to the testimony of the AIB’s expert, Kim A. Scott, who conceded under cross-examination that the lack of balancing on a total limits basis does create an additional amount of subsidy “that is carried into the excess layer” and does not reach other policyholders.

The commissioner does suggest that the Attorney General’s 2004 filing on the BI ILF issue was an improvement on his 2003 effort: “the [Attorney General] has made advances in effectively explaining the reasoning for his framework for calculating and applying BI ILFs .... We do not find the [Attorney General’s] arguments in this area entirely without merit, however, *807and, we encourage the [Attorney General] to further explore and explain these positions in future proceedings.” What “merit” the commissioner finds in the Attorney General’s arguments is unclear from the 2004 decision, especially in light of its summary conclusion that the Attorney General has not demonstrated the existence of either the subsidy imbalance or the BI ILF application mismatch.

See note 13, supra.