This multi-district consumer action alleges a conspiracy by automobile manufacturers to illegally block lower-priced imports from Canada, which is alleged to have inflated the price of new cars sold in America. We granted this interlocutory appeal under Federal Rule of Civil Procedure 23(f) from the district court’s certifications of (1) a nationwide plaintiff class seeking injunctive relief under section 16 of the Clayton Act and Rule 23(b)(2), and (2) a class seeking damages under the antitrust and consumer protection laws of twenty states and Rule 23(b)(3).
Interlocutory appeals from class certification under Rule 23(f) are especially appropriate where the plaintiffs’ theory is novel or where a doubtful class certification results in financial exposure to defendants so great as to provide substantial incentives for defendants to settle non-meritorious cases in an effort to avoid both risk of liability and litigation expense. Tardiff v. Knox County, 365 F.3d 1, 3 (1st Cir.2004); Waste Mgmt. Holdings, Inc. v. Mowbray, 208 F.3d 288, 293 (1st Cir.2000); see also West v. Prudential Sec., Inc., 282 F.3d 935, 937 (7th Cir.2002) (“The effect of a class certification in inducing settlement to curtail the risk of large awards provides a powerful reason to take an interlocutory appeal.”). The defendants assert that the amount at stake in the damages classes alone, counting treble damages, could be as much as $3 billion, and these classes include roughly thirteen million car purchasers. There is no reliable means for measuring the size of the injunctive class, but it is potentially massive.1
By the same token, an erroneous failure to certify a class where individual claims are small may deprive plaintiffs of the only realistic mechanism to vindicate meritorious claims. See, e.g., Tardiff, 365 F.3d at 7 (citing Amchem Prods., Inc. v. Windsor, 521 U.S. 591, 617, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997)). All of these concerns are very much implicated here.
*9We summarize our holdings. We reverse the certification of the injunctive class under the Clayton Act for lack of a live controversy and order dismissal of the claim. Because there is no jurisdiction under the Clayton Act, we remand to the district court for determination of the several issues concerning the existence of federal jurisdiction. On the representation of the parties that there is diversity jurisdiction over at least some of the state damages claims, we review the certification of the damages classes. We vacate that certification; the district court is free to reconsider the class certification orders on a more complete record.
I.
Background
The extensive background of this case is recited in many thoughtful decisions of the district court. See, e.g., In re New Motor Vehicles Canadian Export Antitrust Litig. (Motor Vehicles I), 307 F.Supp.2d 136 (D.Me.2004) (dismissing federal antitrust damages claims); In re New Motor Vehicles Canadian Export Antitrust Litig. (Motor Vehicles II), 335 F.Supp.2d 126 (D.Me.2004) (asserting pendent personal and supplemental subject matter jurisdiction over state law claims); In re New Motor Vehicles Canadian Export Antitrust Litig. (Motor Vehicles III), 350 F.Supp.2d 160 (D.Me.2004) (denying in part motion to dismiss); In re New Motor Vehicles Canadian Export Antitrust Litig. (Motor Vehicles IV), No. MDL 1532, 2006 WL 623591 (D.Me. Mar. 10, 2006) (certifying federal injunctive relief class); In re New Motor Vehicles Canadian Export Antitrust Litig. (Motor Vehicles V), 235 F.R.D. 127 (D.Me.2006) (certifying six exemplar state damages classes); In re New Motor Vehicles Canadian Export Antitrust Litig. (Motor Vehicles VI), 241 F.R.D. 77 (D.Me.2007) (supplemental order on certification of state damages classes); In re New Motor Vehicles Canadian Export Antitrust Litig. (Motor Vehicles VII), 243 F.R.D. 20 (D.Me.2007) (order certifying class action and appointing class counsel); In re New Motor Vehicles Canadian Export Antitrust Litig. (Motor Vehicles VIII), 243 F.R.D. 17 (D.Me.2007) (memorandum order in support of class certification order). Before describing the class certification issues, we sketch the outline of the underlying case.
The plaintiffs’ claim is that from at least 2001 through 2003, the currency exchange rate differential between the strong United States dollar and the cheaper Canadian dollar created arbitrage opportunities2 in the gray market3 to sell lower-priced Canadian cars in the United States. These arbitrage opportunities arose from the difference between the price at which a broker could buy a vehicle in Canada (plus the costs of exporting the vehicle to the United States) and the price at which the broker could resell the vehicle, whether to a dealer or consumer, in the United States. These arbitrage opportunities were enhanced by liberal trade agreements and the harmonization of automotive safety and environmental standards between the two countries in the 1990s. As a result, an exporter could make most new cars sold in *10Canada physically indistinguishable from comparable models sold in the United States by replacing the speedometer and odometer with gauges that measured miles instead of kilometers. All of these circumstances converged in what plaintiffs refer to as a “perfect storm” of cross-border arbitrage opportunities.
Plaintiffs allege that individual automobile manufacturers engaged in business practices, both legal and illegal, designed to restrict the flow of Canadian cars into the United States. Manufacturers allegedly refused to honor warranties on Canadian cars in the United States and discouraged dealers from installing domestic gauges on Canadian cars; they mandated “no export” clauses in sales agreements between dealers and consumers and required Canadian dealers to conduct due diligence into whether potential customers were likely to export their cars out of Canada; and they withheld information about safety recalls from exporting customers. Manufacturers also allegedly imposed disciplinary measures on Canadian dealers who sold to exporting customers. When Canadian cars were discovered in the United States, automakers would impose a “chargeback,” a monetary penalty sometimes amounting to thousands of dollars, on the Canadian dealer who sold the car; manufacturers threatened to withhold inventory of desirable models from offending dealerships; and they threatened to terminate dealerships that sold to exporters.
Plaintiffs allege that these business practices had the effect of suppressing the supply of Canadian cars in the United States. This stifling of supply led to increases in two key prices in the domestic United States automobile market: the Manufacturer’s Suggested Retail Price (“MSRP”) and the dealer invoice price. The MSRP represents the retail price presented to the public. The dealer invoice price represents the manufacturer-determined net wholesale price to dealers.4 Both the MSRP and the list dealer invoice price are determined annually by manufacturers and apply nationally. Under plaintiffs’ theory, these two prices in turn define the bargaining parameters for the price actually paid by consumers. Actual sales prices vary according to individual negotiations between dealers and consumers, but plaintiffs assume that negotiated prices fall within the range between the dealer invoice price and the MSRP.
Defendants5 take the position that the business practices described above amount to vertical restraints between manufacturers and their dealers that do not violate the antitrust laws. Plaintiffs, however, allege that defendants violated section 1 of the Sherman Act by conspiring to coordi*11nate their anti-export business practices across the industry in response to increased arbitrage opportunities after 2000. Plaintiffs allege that defendants accomplished the illegal conspiracy over the course of several meetings and through dissemination of best practices and other information through the dealer associations. These illegal horizontal agreements allegedly enhanced the effectiveness of the extant legal vertical restraints. According to plaintiffs’ theory, “but for” the illegal horizontal conspiracy and its suppression of inter-brand competition, American consumers would have paid lower prices and have thus suffered antitrust-type injury.
For example, the plaintiffs allege that Ford’s 2002 Windstar LX minivan had an MSRP of $16,448 (USD) in Canada and $22,340 (USD) in the United States — a difference of 26%. The MSRP for a 2002 Lexus SC430 was 13% lower in Canada than in the United States. Plaintiffs claim that if the car-makers had not conspired to impede arbitrageurs from exporting Canadian cars to the United States, the manufacturers could not maintain these pricing gaps and instead would have been forced to compete by lowering prices across the United States.
To summarize, plaintiffs’ theory of antitrust impact operates in two stages. In the first stage, the horizontal conspiracy allowed the manufacturers to maintain artificially inflated national MSRPs and dealer invoice prices within the United States. In the second stage, the higher official pricing resulted in higher purchase prices paid by individual consumers.
Throughout the proceedings in the' district court, the defendants have denied any conspiracy. They have also argued that even if such a conspiracy existed, it would not have impacted the domestic market in any case due to the effects of the entirely legal vertical restraints.6 Defendants have also argued that contrary to plaintiffs’ theory, there would have been no flood of Canadian cars into the United States given the number of new cars sold in Canada, the limited opportunities for arbitrage, and the small size and spottiness of the gray market. Defendants assert that even if there were some impact on MSRPs and dealer invoice prices in the American market, there was no widespread antitrust impact on consumers, and plaintiffs have no viable theory to show either causation or damages that would apply to the plaintiff classes.
II.
Nationwide Injunctive Relief Plaintiff Class
Defendants challenge the certification of the injunctive class on the basis that plaintiffs lack standing to sue for injunctive relief under the federal antitrust laws. Because standing is a threshold consideration, the resolution of which may render subsequent arguments irrelevant, we discuss it first.7 Warth v. Seldin, 422 U.S. *12490, 498-99, 95 S.Ct. 2197, 45 L.Ed.2d 343 (1975).
Plaintiffs seek federal injunctive relief pursuant to section 16 of the Clayton Act on behalf of a nationwide class under Rule 23(b)(2) for alleged violations of section 1 of the Sherman Act. Section 16 makes available injunctive relief “against threatened loss or damage by a violation of the antitrust laws.” 15 U.S.C. § 26; see also Cargill, Inc. v. Monfort of Colo., Inc., 479 U.S. 104, 110-11, 107 S.Ct. 484, 93 L.Ed.2d 427 (1986). Plaintiffs allege that injunctive relief is necessary because the conspiracy is ongoing and will continue unless enjoined by the court.
Specifically, plaintiffs seek an injunction that would require defendants to honor warranties in the United States on all new motor vehicles sold in Canada; enjoin the defendants from blacklisting Canadian exporters; enjoin the defendants from exchanging certain information with each other, including blacklists and methods for avoiding export sales; enjoin chargebacks to Canadian dealers; enjoin the tracking of Canadian cars’ Vehicle Identification Numbers (“VINs”); enjoin American manufacturers from penalizing American dealers for buying or selling Canadian vehicles; enjoin the Canadian Automobile Dealers’ Association (“CADA”) and the National Automobile Dealers’ Association (“NADA”) from discouraging or acting to prevent Canadian exports; and enjoin the defendants from withholding safety recall information based on a vehicle’s export status. Motor Vehicles IV, 2006 WL 623591, at *1-2.
On March 10, 2006, the district court issued an order certifying a nationwide injunctive class under Rule 23(b)(2).8 Id. at *1. In doing so, the court noted that a national injunction was the only relief available to members of the proposed class who reside in twenty-seven states because the court had previously dismissed a proposed nationwide class for damages under the federal antitrust laws as well as state-law damages claims for all but twenty-three states and the District of Columbia. Id. at *7. The district court has, with the agreement of the parties, stayed determination of class notice issues pending this appeal.
The district court addressed the scope of standing to seek injunctive relief under section 16 of the Clayton Act in the course of analyzing the typicality requirement of Rule 23(a)(3). Motor Vehicles IV, 2006 WL 623591, at *3-4. Defendants had argued that some named plaintiffs did not have standing to sue for an injunction because not all named plaintiffs actually paid an overcharge due to the alleged conspiracy, and some actually paid less because of it. Id. at *3.
The court correctly held that a plaintiff seeking relief under section 16 need not show actual antitrust damages but only a “threatened loss or damage.” 15 U.S.C. § 26; see also Cargill, 479 U.S. at 111, 107 *13S.Ct. 484; Ocean State Physicians Health Plan, Inc. v. Blue Cross & Blue Shield of R.I., 883 F.2d 1101, 1106 (1st Cir.1989). This in turn means that the requirements for standing for injunctive relief are less stringent than those under section 4 of the Clayton Act, 15 U.S.C. § 15, for standing to pursue damages claims. Cia Petrolera Caribe, Inc. v. Arco Caribbean, Inc., 754 F.2d 404, 407-08 (1st Cir.1985) (“Congress empowered a broader range of plaintiffs to bring § 16 actions because the standards to be met are less exacting than those under § 4; under § 16, a plaintiff need show only a threat of injury rather than an accrued injury.”).
Plaintiffs must demonstrate “a significant threat of injury from an impending violation ... or from a contemporary violation likely to continue or recur.” Mid-West Paper Prods. Co. v. Cont’l Group, Inc., 596 F.2d 573, 591 (3rd Cir.1979) (quoting Zenith Radio Corp. v. Hazeltine Research, Inc., 395 U.S. 100, 130, 89 S.Ct. 1562, 23 L.Ed.2d 129 (1969)) (internal quotation marks omitted). The threatened injury must be an injury for which the plaintiff would be entitled to compensation if the injury actually occurred. Cargill, 479 U.S. at 110-13, 107 S.Ct. 484. There is no doubt that the types of injuries alleged here — consumers paying artificially inflated prices due to antitrust violations — are antitrust injuries. Reiter v. Sonotone Corp., 442 U.S. 330, 339, 99 S.Ct. 2326, 60 L.Ed.2d 931 (1979); see also Motor Vehicles IV, 2006 WL 623591, at *3 (“Higher prices for consumers resulting from antitrust violations are antitrust injuries.”).
The district court rejected the defendants’ standing argument, reasoning that whether or not the named plaintiffs suffered an actual overcharge, “[t]hey confront or confronted a threatened loss or damage resulting from restrictions on competition ... [that] the antitrust laws were designed to prevent.” Id. at *4. The district court relied on two propositions that are now called into question. First, the court reasoned that “whether exchange rates during some periods [within the class period] temporarily made Canadian cars more expensive and therefore not price competitive, does not affect the standing of these plaintiffs to seek an injunction against continuation of such a conspiracy.” 9 Id. (emphasis added). Second, the court noted in an opinion now two years old that “[t]here is no indication that exchange-related arbitrage opportunities have permanently ended.” Id.
As the defendants point out on appeal, the standing issue emanates from Article Ill’s requirement that there remain a live case or controversy throughout the course of a litigation. Golden v. Zwickler, 394 U.S. 103, 108-09, 89 S.Ct. 956, 22 L.Ed.2d 113 (1969). If a named plaintiff fails to establish such a continuing controversy, she normally may not invoke the power of the federal courts to seek relief on her own behalf or that of other members of a putative class. O’Shea v. Littleton, 414 U.S. 488, 494, 94 S.Ct. 669, 38 L.Ed.2d 674 (1974); see also Sosna v. Iowa, 419 U.S. 393, 402, 95 S.Ct. 553, 42 L.Ed.2d 532 (1975) (“There must not only be a named plaintiff who has ... a case or controversy at the time the complaint is filed, and at *14the time the class action is certified by the District Court ... but there must be a live controversy at the time this Court reviews the case.”)- Under certain circumstances, a court may find that the requisite controversy exists “between a named defendant and a member of the class represented by the named plaintiff, even though the claim of the named plaintiff has become moot” due to that plaintiffs individual circumstances. Sosna, 419 U.S. at 402, 95 S.Ct. 553.
The defendants contend that even if the named plaintiffs had standing to seek in-junctive relief at some time in the past, circumstances have changed such that there is no longer a live controversy between named plaintiffs and the defendants to bolster plaintiffs’ standing to seek in-junctive relief. We agree.
Section 16’s requirement of “threatened injury,” 15 U.S.C. § 26, dovetails with Article Ill’s requirement that in order to obtain forward-looking relief, a plaintiff must face a threat of injury that is both “ ‘real and immediate,’ not ‘conjectural’ or ‘hypothetical.’ ” O’Shea, 414 U.S. at 494, 94 S.Ct. 669 (quoting Golden, 394 U.S. at 109-10, 89 S.Ct. 956). “Past exposure to illegal conduct does not in itself show a present case or controversy regarding in-junctive relief ... if unaccompanied by any continuing, present adverse effects.” Id. at 495-96, 89 S.Ct. 956.
As with the other elements of standing, it is plaintiffs who, by invoking federal jurisdiction, bear the burden of establishing the presence of a threat of injury. McInnis-Misenor v. Me. Med. Ctr., 319 F.3d 63, 67 (1st Cir.2003) (citing Bennett v. Spear, 520 U.S. 154, 167-68, 117 S.Ct. 1154, 137 L.Ed.2d 281 (1997)). “There must be some immediacy or imminence to the threatened injury.” Id. at 68 (citing City of Los Angeles v. Lyons, 461 U.S. 95, 101-02, 103 S.Ct. 1660, 75 L.Ed.2d 675 (1983)).
Plaintiffs have failed to establish the continuing presence of the requisite threatened injury. The “perfect storm” that allegedly precipitated massive arbitrage opportunities for selling Canadian cars in the United States ceased long ago. Plaintiffs’ theory of antitrust violation posits that exceptional arbitrage opportunities arose early in this decade due to a combination of relaxed trade restrictions between the United States and Canada, physical harmonization of cars manufactured for the two markets, and a differential between the values of American and Canadian currencies. The last factor has varied since the beginning of the class period, and plaintiffs themselves measure windows of actual arbitrage opportunity in large part according to undulations in exchange rates.
After the district court certified the in-junctive class, it ordered the parties to make submissions regarding proposed end dates for the damages classes. Plaintiffs’ expert submitted an affidavit indicating that arbitrage opportunities for Canadian cars significant enough to impact United States prices for new vehicles ended in May 2003. The district court accordingly adopted April 30, 2003 as the endpoint for the state damages classes. Motor Vehicles VI, 241 F.R.D. at 79. Plaintiffs have not made any effort since the district court’s ruling to demonstrate whether or when prevailing conditions — primarily, exchange rates between Canada and the United States — might reasonably be expected once again to resemble those during the class period. This is for good reason, as the value of the United States dollar relative to the Canadian dollar in the years 1998-2003 hit an apex unseen in at least *15the last half century.10 In the nearly five years since April 2003, the time at which plaintiffs concede actual injury to the damages classes abated, the exchange rate has fallen in favor of the Canadian dollar.11 In the circumstances of this case, that alone eliminates any realistic current threat.
The speculative nature of the claim that exchange rates could one day create additional arbitrage opportunities is reinforced by a second contingency inherent in the plaintiffs’ theory of threatened future harm. In order for antitrust injury to befall a class member, she must also intend to purchase a new car at a time when arbitrage opportunities are ripe for exploitation. The district court certified an in-junctive class defined as “[a]ll persons ... who purchased or leased or intended to purchase or lease a new motor vehicle manufactured by a defendant from a United States dealer during the period from January 1, 2001, to March 10, 2006.” The complaint names thirty-six individuals who bought or leased new cars in the United States during the class period. Nowhere, however, does the complaint make any allegation regarding a named plaintiffs intention to buy or lease another new vehicle within such a time frame as could be deemed imminent. The fact that the in-junctive class has been restricted to a period of time covering 2001 to early 2006 underscores the speculative nature of the plaintiffs’ hypothetical future injury.
In McInnis-Misenor v. Maine Medical Center, this court inquired whether the plaintiff faced an “immedia[te] or imminen[t]” threat of injury that would support standing in federal court. 319 F.3d at 68. We held that standing was lacking where any possible injury to the plaintiff was “contingent on several events which may or may not happen.” Id. at 72. Here, there is nothing to suggest that any named plaintiff harbors an “imminent” intention to buy a new car in coincidence with another “perfect storm” of arbitrage-friendly market conditions.
Plaintiffs attempt to revive the controversy by arguing that the defendants’ conspiracy to hinder cross-border sales is ongoing. Even if this were true, the argument ignores the point that the relevant question is whether plaintiffs suffer a realistic threat of injury stemming from those alleged antitrust violations. An antitrust violation and standing are distinct and independently necessary prerequisites for the relief sought by the injunctive class. See II Areeda & Hovenkamp, Antitrust Law ¶ 335f, at 297 (2d ed.2000); cf. SAS of P.R., Inc. v. P.R. Tel. Co., 48 F.3d 39, 43 (1st Cir.1995).
*16The Rule 23(b)(2) class is vacated for lack of a live controversy between the parties such as would justify an injunctive remedy. The injunctive relief claim is dismissed.
III.
Jurisdiction over Remaining Plaintiff Classes
The claim for injunctive relief under the Clayton Act provided the federal jurisdictional spine of this case. That basis for federal jurisdiction is now gone, and the question of federal jurisdiction over the state-law damages claims presents itself. It may be that the district court could retain jurisdiction over the state damages claims on either of two theories, but we leave it to the district court to decide on remand whether there is jurisdiction.12
The parties and the district court have referred to diversity, see 28 U.S.C. § 1332,13 as a basis for maintaining the state damages classes in federal court. To this point in the proceedings, however, the state damages claims have been analyzed as within the court’s supplemental jurisdiction based on the presence of the federal claim. See, e.g., Motor Vehicles II, 335 F.Supp.2d 126.14 Whether the diversity basis is sound has not been tested.
At oral argument, the defendants pointed out that proceeding on the basis of diversity jurisdiction would require analysis of each individual state-based class. Defendants also conceded that diversity could provide an independent basis for federal jurisdiction for at least some of the state-law damages classes. The district court has not had the opportunity to assess whether each state class satisfies the requirements of § 1332 and may do so on remand.
The district court may, in the alternative, consider whether to exercise its discretion to continue exerting supplemental jurisdiction, see 28 U.S.C. § 1367, over the state damages claims. See Rodríguez v. Doral Mortgage Corp., 57 F.3d 1168, 1177 (1st Cir.1995) (“In an appropriate situation, a federal court may retain [supplemental] jurisdiction over state-law claims notwithstanding the early demise of all foundational federal claims.”). In weighing this option, the district court should consider “the totality of the attendant circumstances,” including considerations of judicial economy, fairness to the parties, and the nature of the applicable state law. Id.
*17On the assumption that there is some basis for federal jurisdiction over at least some remaining state class claims, we turn to the certification of the state damages classes.
IV.
Standards of Review of Class Certification Decisions
We review class certification rulings, including those for damages classes, for abuse of discretion. Mowbray, 208 F.3d at 295. This standard has teeth:
An abuse occurs when a court, in making a discretionary ruling, relies upon an improper factor, omits consideration of a factor entitled to substantial weight, or mulls the correct mix of factors but makes a clear error of judgment in assaying them. An abuse of discretion also occurs if the court adopts an incorrect legal rule.
Id. (citation omitted). Furthermore, a class certification appeal “can pose pure issues of law reviewed de novo” Tardijf, 365 F.3d at 4. Review of mixed questions of law and fact varies from non-deferential review for law-dominated issues to deferential clear-error review for fact-dominated ones. In re PolyMedica Carp. Sec. Litig. (PolyMedica), 432 F.3d 1, 4 (1st Cir.2005) (citing Johnson v. Watts Regulator Co., 63 F.3d 1129, 1132 (1st Cir.1995)).
Intertwined with the scope of our review on appeal is the question of how far a district court should go in testing legal and factual premises at the class certification stage. When such premises are disputed, the court may “probe behind the pleadings,” Gen. Tel. Co. of Sw. v. Falcon, 457 U.S. 147, 160, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982), to “formulate some prediction as to how specific issues will play out” in order to assess whether the proposed class meets the legal requirements for certification, Mowbray, 208 F.3d at 298. In short, “a court has the power to test disputed premises [at the certification stage] if and when the class action would be proper on one premise but not another.” Tardiff, 365 F.3d at 4-5. Here, the premises are challenged as to both steps of plaintiffs’ theory.
In performing this predictive function and in “[e]xercising its broad discretion, ... the district court must evaluate the plaintiffs evidence ... critically without allowing the defendant to turn the class-certification proceeding into an unwieldy trial on the merits.” PolyMedica, 432 F.3d at 17. However, as both we and other circuit courts have emphasized, weighing whether to certify a plaintiff class may inevitably overlap with some critical assessment regarding the merits of the case. See, e.g., Regents of the Univ. of Cal. v. Credit Suisse First Boston (USA), Inc., 482 F.3d 372, 380 (5th Cir.2007), cert. denied, — U.S.-, 128 S.Ct. 1120, 169 L.Ed.2d 957 (2008). It would be contrary to the “rigorous analysis of the prerequisites established by Rule 23 before certifying a class” to put blinders on as to an issue simply because it implicates the merits of the case. Smilow v. Sw. Bell Mobile Sys., Inc., 323 F.3d 32, 38 (1st Cir.2003). Our review of a Rule 23(f) appeal necessarily includes “review [of] the merits of -thé district court’s theory of liability insofar as they also concern issues relevant to class certification.” Credit Suisse, 482 F.3d at 381.
V.
State Damages Classes
A. Procedural History
Applying Illinois Brick v. Illinois, 431 U.S. 720, 97 S.Ct. 2061, 52 L.Ed.2d 707 (1977), the district court dismissed the *18plaintiffs’ federal antitrust damages claims in March 2004 because plaintiffs were indirect purchasers. Motor Vehicles I, 307 F.Supp.2d at 137. Illinois Brick embodies a policy judgment that the reach of damages for federal antitrust violations does not extend to indirect purchasers, even if they have suffered some impact in fact. See II Areeda & Hovenkamp, supra, ¶ 395, at 554-55.
After dismissal of the federal damages claim, plaintiffs filed a second amended complaint which added damages claims under common and statutory law for all fifty states and the District of Columbia. Motor Vehicles III, 350 F.Supp.2d at 168, 207. After considering for each disputed state15 whether plaintiffs had pled facts sufficient to state a claim under that state’s laws, the district court dismissed the state law damages claims for all but twenty-three states and the District of Columbia. Id. at 168. The propriety of that dismissal is not now before us. This appeal is from three orders that did certify state damages classes.
Before discovery was completed, the plaintiffs moved on July 29, 2005, to certify six exemplar classes of state damages claims.16 On May 12, 2006, the court preliminarily certified five of the six exemplar state damages classes under Rule 23(b)(3) and asked for additional materials regarding the date on which the class period should end. Motor Vehicles IV, 235 F.R.D. at 129.17 On March 21, 2007, the court issued a supplemental order concluding that the class period for the state damages claims should end on April 30, 2003, and certifying fifteen additional state damages classes for a total of twenty. Motor Vehicles VI, 241 F.R.D. at 79, 84. On June 15, 2007, the court issued an order and explanatory memorandum under Rule 23(c)(1)(B) defining the class and class claims, issues, and defenses. Motor Vehicles VII, 243 F.R.D. 17; Motor Vehicles VIII, 243 F.R.D. 20.
1. Requirements for Class Certification
To certify the statewide damages classes, the district court had to evaluate whether the four threshold requirements of Rule 23(a) were met, as well as whether Rule 23(b)(3)’s two additional prerequisites were satisfied. Amchem, 521 U.S. at 614, 117 S.Ct. 2231; Smilow, 323 F.3d at 38. Rule 23(a) requires that (1) there be nu-merosity, (2) there be common questions of law or fact, (3) the class representative’s claims be typical of the class, and (4) the representative’s representation of the class be adequate. Fed.R.Civ.P. 23(a). To certify a class under Rule 23(b)(3), a judge must further find “that the questions of law or fact common to class members predominate over any questions affecting only individual members” (“predominance”), and that “a class action is superior to other available methods for fairly and efficiently adjudicating the controversy” (“superiority”). Fed.R.CivJP. 23(b)(3). In classes certified under Rule 23(b)(3), the Rules “invite[ ] a close look at the case before it is accepted as a class action.” Amchem, 521 U.S. at 615, 117 S.Ct. 2231 (quoting B. *19Kaplan, Continuing Work of the Civil Committee: 1966 Amendments of the Federal Rules of Civil Procedure (I), 81 Harv. L.Rev. 356, 390 (1967)).
The court held that numerosity was not disputed and that adequacy was established. Motor Vehicles V, 235 F.R.D. at 130, 141. It also concluded that a class action was by far the superior method for proceeding. Id. at 148. The real dispute revolved around whether common evidence could be used to prove the impact of the alleged conspiracy on U.S. consumers (“common impact”) and any resulting damages (“common proof of damages”). Id. at 129, 132.18 The court noted it did not yet have enough evidence to determine the merits question of whether the plaintiff classes actually suffered antitrust or consumer protection injury. It repeatedly emphasized it would address that question later on a proper record — for example, at summary judgment.19 Id. at 131, 136, 139.
As the district court noted, there is some overlap among the certification criteria of commonality, Rule 23(a)(2), typicality, Rule 23(a)(3), and predominance, Rule 23(b)(3). Id. at 130 n. 4 (“[I]f the proof of impact is not common across the class, then not only is the named plaintiffs’ claim of injury not typical, but the predominance assessment is also affected.”); see also Amchem, 521 U.S. at 623 n. 18, 117 S.Ct. 2231 (predominance and typicality are similar in some respects); 6 A. Conte & H.B. Newberg, Newberg on Class Actions § 18:8, at 24 n. 2 (4th ed.2002). The questions in this case of common impact (antitrust-type causation) and common proof of damages are relevant to all three criteria.
Rule 23(a)’s requirement of commonality is a low bar, and courts have generally given it a “permissive application.” 7A Charles Alan Wright, Arthur R. Miller, Mary Kay Kane, Federal Practice and Procedure § 1763, at 221 (3d ed.2005). The district court did not have trouble finding sufficient commonality in this case. The court pointed to the common questions of whether there was a conspiracy; if so, whether it affected either dealer invoice prices or MSRPs; and whether there was any violation of state antitrust or consumer protection laws. Motor Vehicles V, 235 F.R.D. at 130; see also 6 Conte & New-berg, supra, § 18:5, at 14 & n. 3 (gathering cases that hold that “allegations concerning the existence, scope, and efficacy of an alleged conspiracy present questions ade*20quately common to class members to satisfy the commonality requirement”).
“[T]he predominance criterion is far more demanding,” however, than the commonality requirement. Amchem, 521 U.S. at 624, 117 S.Ct. 2231. Under the predominance inquiry, “a district court must formulate some prediction as to how specific issues will play out in order to determine whether common or individual issues predominate in a given case.” Mowbray, 208 F.3d at 298. In antitrust class actions, common issues do not predominate if the fact of antitrust violation and the fact of antitrust impact cannot be established through common proof. See Blades v. Monsanto Co., 400 F.3d 562, 566 (8th Cir.2005). The district court did not find predominance and typicality as easily established as commonality. The court chose to analyze under the heading of typicality whether plaintiffs were asserting sufficiently common proof of impact, Motor Vehicles V, 235 F.R.D. at 130-40, and it considered under the heading of predominance whether any resulting damages would likewise be established by sufficiently common proof, id. at 142-45. In consideration of these issues, the district court relied on the submissions of the parties’ experts.
2. Submissions of Expert Witnesses
Plaintiffs relied primarily on their expert Professor Robert E. Hall of Stanford University and the Hoover Institute, who, along with the defense expert, assumed that plaintiffs’ allegations of conspiracy were true. In his July 2005 report, Professor Hall noted that collectively the defendants’ sales accounted for 89% of the U.S. market and 84% of the Canadian market. He concluded that “if defendants were unable to impose export restraints or were less effective in imposing them, they would [have] lower[ed] U.S. [dealer] invoice prices and MSRPs for most, if not all, of their vehicles.” Expert Report of Robert E. Hall, Ph.D. ¶ 51. Even though retail sales of cars are individually negotiated, Professor Hall opined that class members would have experienced common impact from the changed MSRPs and dealer invoice prices because a change in these prices would shift the entire negotiating range, benefiting (or harming) essentially all consumers.
Professor Hall initially proposed two different approaches to establishing common evidence of damages to class members. The first approach would be to rely on statistical models (“Nash equilibri-ums”) used in the auto industry to predict market outcomes. Under his second approach, a benchmark method, the U.S.Canadian market would be evaluated against a comparable market not affected by the challenged conduct. Both approaches, Professor Hall asserted, would account for “heterogeneity across vehicles, dealers, consumers, and time, [and] they can be used to identify cars or time periods for which there are no damages.” Id. ¶ 62. Professor Hall also asserted that data existed that would allow him to implement these models, though the most detailed data were in the hands of defendants.
The defense expert, Professor Joseph P. Kalt of Harvard University, filed his expert report on September 30, 2005. Even assuming a conspiracy, he disputed that there was any common impact on either the national manufacturer-to-dealer prices or the individual dealer-to-consumer sales. His critique focused on a description of the actual gray market for Canadian vehicles in the United States, which he asserted was spotty, erratic, and too insignificant to affect the national market, even absent any collusive activity. He also criticized Professor Hall for not distinguishing be*21tween the effects of the manufacturers’ legal vertical restraints and those of the alleged horizontal conspiracy. As to damages, he disputed that there were any common methods of proof and asserted that Professor Hall’s suggested approach of employing Nash equilibriums was just “an empty phrase.” Expert Report of Joseph P. Kalt, Ph.D. 63-64.
In reply to Professor Kalt’s critique, Professor Hall submitted a rebuttal report on December 20, 2005. He took issue with Professor Kalt’s conclusions regarding the gray market20 and, applying regression analysis to Professor Kalt’s data on arbitrage opportunity and export activity, concluded that “a larger price gap [between Canadian and United States car prices] is associated with greater export activity and that this relationship is not the result of randomness.” Rebuttal Report of Robert E. Hall, Ph.D. ¶25. Specifically, his regressions suggested that “a $1,000 increase in the price gap for a given model [of car] is, on average, associated with an increase in exports of at least 33 percent.” Id. ¶ 24. The minor increase in gray market sales in 2001-2002 despite the conceded greater arbitrage opportunities, then, did not indicate the lack of a conspiracy, but rather suggested an effective one.
Professor Hall argued that a Nash equilibrium model could distinguish the effects of unilateral restraints from those of any horizontal conspiracy, and he also suggested that further discovery might reduce the need for modeling by yielding direct evidence on the effectiveness of the unilateral and alleged horizontal constraints. He stated that he had offered preliminary mathematical formulas at his deposition to establish the viability of both this approach and the approach of his proposed damages model, and he included the preliminary damages model in his rebuttal report. Professor Hall again asserted his view that there were methods based on common evidence, such as the econometric model he was proposing, that could account for the “major dimensions in which damages might vary among potential class members.” Id. ¶ 50.
3. Typicality and Predominance
On this evidence, the district court preliminarily certified five exemplar state classes. The court held that the presentation by Professor Hall, supported by extrinsic economic studies, sufficed for purposes of showing common proof of impact.21 The court found “unexceptionable” the plaintiffs’ theory that other things being equal, a restriction on the supply of lower-priced cars coming into the United States market will exert an upward pressure on domestic car prices. Motor Vehicles V, 235 F.R.D. at 137. This pressure will apply both to the prices dealers pay to manufacturers and to the prices paid by consumers to dealers. While individual negotiations may determine the final price, the starting point for most negotiations is the MSRP, and the final purchase price for most consumers is between the dealer invoice price and the MSRP. Id. In effect, the *22overall negotiating range would be elevated, resulting in higher consumer prices across the board. Id. at 138-39. The court accepted some of these contentions as adequate to demonstrate that the named plaintiffs’ claims would be typical of the class, although the court was careful to note that it was not then deciding whether plaintiffs’ proof of impact was sufficient to withstand a motion for summary judgment. Id. at 139.
The court was also careful to note that the proffered common proof of impact might be insufficient under some states’ laws. Id. at 132. The court sua sponte examined the laws of five of the exemplar states to determine what level of proof of consumer impact each state required and what inferences were acceptable to show impact.22 The court concluded that there was a range. California, at one end, permitted “an inference of antitrust impact, even as to indirect purchasers, from the existence of the conspiracy.”23 Id. at 135; see also id. (“In the consumer context, at least a portion of the illegal overcharge by a manufacturer will presumably be passed on by the independent distributors to consumer class members in the form of higher prices.” (quoting Global Minerals & Metals Corp. v. Superior Court, 113 Cal.App.4th 836, 7 Cal.Rptr.3d 28, 44-45 (2003)) (internal quotation marks omitted)).
Maine was at the other end of the spectrum, requiring evidence, not inference, of impact. While the Maine antitrust statute explicitly permits recovery for indirect injury, Me.Rev.Stat. Ann. tit. 10, § 1104, the Maine Law Court has not yet commented on what indirect purchasers must show to establish impact or causation. Motor Vehicles V, 235 F.R.D. at 132. The district court turned to Maine superior court decisions, which the court summarized as holding that “indirect purchasers in Maine must produce specific proof that they paid higher prices as a result of the conspiracy (in the face of the possibility that all such increases were absorbed at the retailer level).” Id. at 134; see also id. (“Because indirect purchasers must demonstrate that overcharges have been passed on to them, such claims present an entirely separate level of evidence and proof than that found in a direct purchaser claim.” (quoting Melnick v. Microsoft Corp., Nos. CV-99-709, CV-99-752, 2001 WL 1012261, at *6 (Me.Super.Ct. Aug. 24, 2001))). Likewise, the Maine consumer protection statute, Me.Rev.Stat. Ann. tit. 5, §§ 205-A to 214, allows indirect purchasers to recover, but injury is not presumed. Motor Vehicles V, 235 F.R.D. at 135 (citing State v. Weinschenk, 868 A.2d 200 (Me.2005)).
The court also discussed the states that fell between those two poles. New Mexico, it concluded, seems to allow indirect purchasers to establish antitrust impact through correlation analysis. See id. at 136. Tennessee and Vermont have not yet fully addressed the question. See id. The court noted that in states like Maine where the passing on of an antitrust or consumer protection injury to indirect purchasers cannot be presumed, plaintiffs might have difficulty proving injury to individual consumers if their common proof can establish only an inference of injury. See id. at 139.
Turning to the question of predominance under Rule 23(b)(3), the court found there were at least five common disputed issues *23weighing in favor of class certification.24 Id. at 142. Defendants argued that the lack of common impact at both stages of plaintiffs’ theory and the lack of common proof of damages defeat predominance. The court reiterated that it had found plaintiffs’ proposal to prove impact through common proof preliminarily sufficient, and it pointed out that the existence of a common disputed issue weighs in favor of class certification, not against it. Id. at 142 (citing Tardiff, 365 F.8d at 4-5). The adequacy of plaintiffs’ proof of common impact, and whether the impact even constitutes cognizable antitrust or consumer protection injury, “are merits determinations that are common in each [state] class” and would be resolved at trial. Id
The court also disagreed that the question of individual damages defeats the predominance of common questions. First, the court noted that “[w]here, as here, common questions predominate regarding liability, then courts generally find the predominance requirement to be satisfied even if individual damages issues remain.” Id. at 143 (quoting Smilow, 323 F.3d at 40). Second, the court stated that it would not determine at the class certification stage whether plaintiffs’ method of proving damages was adequate. Id. at 144. The mere fact that there are differences among members of a class regarding their individual amounts of damages does not preclude class certification. Smilow, 323 F.3d at 40 (“The individuation of damages in consumer class actions is rarely determinative under Rule 23(b)(3).”); 7B Charles Alan Wright, Arthur R. Miller, Mary Kay Kane, supra, § 1781, at 235. Often those variations can be determined according to a universal mathematical or formulaic calculation, obviating the need for evidentiary hearings on each individual claim. Smi-low, 323 F.3d at 40. Plaintiffs proposed to use such an approach here. Motor Vehicles V, 235 F.R.D. at 143-44. The district court noted that it was “not overwhelmed by the plaintiffs’ offer of damage calculation models,” but it “conclude[d] that damages [were] not yet a ground to deny certification.” Id. at 144-^5; see also id. at 145 n. 63 (specifying some of the problems plaintiffs’ damages models faced).
Finally, the district court raised the question of the end date for the damages classes. Id. at 140. Another round of expert reports ensued, accompanied by briefing repeating many of the same arguments. The court focused on the choice of end date in its March 21, 2007 order; it commented that the defendants’ additional critiques did not alter its views that a class should be certified.25 Motor Vehicles VI, 241 F.R.D. at 80-82. Over defendants’ *24insistence that the court determine whether the “alleged horizontal conspiracy actually impacted American car prices,” the court refused to reexamine the issue — as well as the related issue of “what vertical restraints the individual manufacturers maintained, their legality and their effect” — at that time. Id. at 80-82 & n. 5. On June 15, 2007, the court entered an order of class certification in compliance with Rule 23(c)(1)(B).
B. Defendants’ Appeals from Certification of the Damages Classes
In challenging the certification of the state damages classes, defendants primarily argue that the district court did not engage in a sufficiently searching inquiry into the relevant merits issues. It is a settled question that some inquiry into the merits at the class certification stage is not only permissible but appropriate to the extent that the merits overlap the Rule 23 criteria. Falcon, 457 U.S. at 160, 102 S.Ct. 2364; PolyMedica, 432 F.3d at 6; Mowbray, 208 F.3d at 297-98. It is less settled what degree of merits inquiry is required at the class certification stage, and the Supreme Court has not yet addressed the issue.
Our sister circuits agree that when class criteria and merits overlap, the district court must conduct a searching inquiry regarding the Rule 23 criteria, but how they articulate the necessary degree of inquiry ranges along a spectrum which suggests substantial differences. The Second, Fourth, Fifth, and Seventh Circuits coalesce around the more rigorous end of this spectrum, forbidding district courts from relying on plaintiffs’ allegations of sufficiently common proof and requiring the courts to make specific findings that each Rule 23 criterion is met. Miles v. Merrill Lynch & Co. (In re Initial Pub. Offering Sec. Litig.), 471 F.3d 24, 33, 41 (2d Cir.2006) [hereinafter In re IPO ] (requiring a “definitive assessment of Rule 23 requirements,” including the resolution of relevant factual disputes); Unger v. Amedisys Inc., 401 F.3d 316, 321-22 (5th Cir.2005) (requiring courts to find facts favoring class certification through the use of “rigorous, though preliminary, standards of proof’); Gariety v. Grant Thornton, LLP, 368 F.3d 356, 366 (4th Cir.2004) (requiring that “the factors spelled out in Rule 23 ... be addressed through findings”); Szabo v. Bridgeport Machs., Inc., 249 F.3d 672, 675-76 (7th Cir.2001) (requiring “whatever factual and legal inquiries are necessary under Rule 23” to “resolve the disputes before deciding whether to certify the class”). These circuits’ use of the term “findings” in this context should not be confused with binding findings on the merits. The judge’s consideration of merits issues at the class certification stage pertains only to that stage; the ultimate factfinder, whether judge or jury, must still reach its own determination on these issues. In re IPO, 471 F.3d at 39; Gariety, 368 F.3d at 366.
On the other end of the spectrum, the Third and Eighth Circuits sometimes require an inquiry into and preliminary resolution of disputes, but they do not require findings and do not hold that such inquiry will always be necessary. Blades, 400 F.3d at 567, 575 (holding that sometimes courts will be required to resolve factual disputes preliminarily at the class certification stage but voicing caution); Newton v. Merrill Lynch, Pierce, Fenner & Smith, Inc., 259 F.3d 154, 166 (3d Cir.2001) (“A class certification decision requires a thorough examination of the factual and legal allegations.”); id. at 168 (“In reviewing a motion for class certification, a preliminary inquiry into the merits is sometimes necessary to determine whether the alleged *25claims can be properly resolved as a class action”).
This court has grappled with this issue as well. We have said in Smilow that “a district court must conduct a rigorous analysis” of Rule 23’s prerequisites, 323 F.3d at 38, and in Mowbray that “a district court must formulate some prediction as to how specific issues will play out,” 208 F.3d at 298. See also Tardiff, 365 F.3d at 4-5 (noting that the common presumption at early stages of litigation that “the complaint’s allegations are necessarily controlling” does not apply in class certification situations because “class action machinery is expensive and in our view a court has the power to test disputed premises early on if and when the class action would be proper on one premise but not another”).
In PolyMedica, a securities class action, we said that the court “was entitled to look beyond the pleadings in its evaluation” of both the question of class certification generally and the applicability of the fraud-on-the-market presumption specifically, even though those questions overlapped with the merits of the case. PolyMedica, 432 F.3d at 6. The fraud-on-the-market presumption in securities class actions allows plaintiffs to establish the necessary element of reliance through common proof. If the market is efficient, the theory goes, market prices will incorporate all publicly available information, including material misrepresentations, so that an investor who buys or sells stock in reliance on the integrity of the market price is in fact buying or selling stock in reliance on the material misrepresentations. See id. at 7-8. Because common issues would no longer predominate in a securities class action if plaintiffs could not at trial establish reliance through the common proof of the fraud-on-the-market presumption, courts at the class certification stage probe the factual basis of the fraud-on-the-market presumption to make sure it will be a viable form of proof in a given case.
In both PolyMedica and its companion case, Stuebler v. Xcelera.Com (In re Xcel-era.com Sec. Litig.), 430 F.3d 503 (1st Cir.2005), we, along with the district court, rigorously tested the evidence submitted by both sides to determine whether the fraud-on-the-market presumption was reasonably applicable, specifically whether the plaintiffs would be able to demonstrate that the market was efficient. At the class certification stage, we noted, plaintiffs needed to present “basic facts” that the fraud-on-the-market presumption could be invoked, even though its actual applicability was to be resolved at trial. PolyMedi-ca, 432 F.3d at 19. Our review of the district court’s determination of whether or not the fraud-on-the-market presumption could be invoked was based not on the level of detail in the district court’s explanation, but on “whether the evidence supports its determination to apply the presumption.” Xcelera, 430 F.3d at 512. In Xcelera, for example, the district court actively evaluated the testimony of two competing experts and preliminarily credited the plaintiffs’ expert, a determination this court upheld — after surveying the expert testimony ourselves — on clear error review. Id. at 512-16.
PolyMedica and Xcelera could be read, but we think not properly, as limiting this requirement that district courts probe into the viability of the premises of plaintiffs’ theory of injury to cases employing only legal presumptions of injury. Under this circuit’s approach, in our view, a searching inquiry is in order where there are not only disputed basic facts, but also a novel theory of legally cognizable injury. Plaintiffs cannot make their case without common proof of causation, and they can only prove causation through common means if their novel theory is viable; that viability *26in turn depends on their ability to establish — -whether through mathematical models or further data or other means — the key logical steps behind their theory. Such reliance on a novel theory to establish a primary element of a claim necessitates a more searching inquiry into whether plaintiffs will be able to prove the pivotal elements of their theory at trial. This is especially so when a case implicates the sort of factors that we have deemed important in the Rule 23(f) calculus, namely, when the granting of class status “raises the stakes of litigation so substantially that the defendant likely will feel irresistible pressure to settle.” Mowbray, 208 F.3d at 293.
We do not need to resolve now whether “findings” regarding the class certification criteria are ever necessary, but we do hold that when a Rule 23 requirement relies on a novel or complex theory as to injury, as the predominance inquiry does in this case, the district court must engage in a searching inquiry into the viability of that theory and the existence of the facts necessary for the theory to succeed.
Contrary to defendants’ assertions, the district court did not believe itself limited by Eisen v. Carlisle & Jacquelin, 417 U.S. 156, 94 S.Ct. 2140, 40 L.Ed.2d 732 (1974), to non-merits inquiries. Instead, the court attempted to meet its obligations under PolyMedica, Mowbray, and our other cases to conduct a rigorous analysis at the certification stage. See Motor Vehicles V, 241 F.R.D. at 81 n. 7. The court’s ability to probe into the viability of plaintiffs’ proffered theory and to formulate some predictions as to how key issues in this novel and complex case would develop was hampered, however, by the incomplete record at the time, as well as by the fact that plaintiffs’ expert had not yet fully formulated all aspects of his analysis. The court pointed out that it was ruling on class certification before discovery was completed and was relying upon the plaintiffs’ representation that they would fill in the gaps in their evidence with further discovery and further work. It repeatedly said that it was willing to take another look at these questions when the record was more complete.26
This court has had little occasion to discuss the timing of the issuance of class certification orders, much less of damages classes under Rule 23(b)(3). See Mowbray, 208 F.3d at 299 n. 7. Rule 23(c)(1)(A) says only that the court must act “at an early practicable time.” Rule 23(c)(1)(C) also provides that a class certification order may be altered or amended before final judgment.27 A district court which has taken an initial look at the merits is not foreclosed from later entertaining, post-discovery, a summary judgment motion from defendants asserting that plaintiffs cannot establish the requisite antitrust and consumer protection impact through common means. See 3 Conte & Newberg, swpra, § 7:15, at 48-57. Indeed, it is not uncommon to defer final decision on certifications pending comple*27tion of relevant discovery. Id. § 7:16, at 57-59.
When the decision on class certification is made before full class discovery has been completed, as here, it is necessarily more predictive. As the Eighth Circuit has noted, the decision may require revisiting upon completion of full discovery. Blades, 400 F.3d at 567; see also Gariety, 368 F.3d at 368.
In another case, this posture of certification being decided before completion of class discovery might not raise any concerns. In this case it does because of the novelty and complexity of the theories advanced and the gaps in the evidence proffered. The district court expressed multiple times its concern about the adequacy of several of plaintiffs’ showings and expressed a willingness to revisit the question once it had a better record in front of it. We share those concerns.
Plaintiffs’ theory of impact on indirect purchasers is both novel and complex. Injury in price-fixing cases is sometimes not difficult to establish. Plaintiffs do not, however, advance such a price-fixing theory. Rather, the plaintiffs’ theory is that the higher prices are the result of a “but-for” world. In step one of plaintiffs theory, but for the defendants’ illegal stifling of competition, the manufacturers would have had to set dealer invoice prices and MSRPs lower to avoid losing sales to the lower-priced Canadian cars coming across the border for resale in the United States. In step two, the higher dealer invoice prices and MSRPs enabled by this stifling of competition resulted in injury to consumers in the form of higher retail prices.
The first step of plaintiffs’ theory requires demonstrating that the defendants’ actions did result in an increase in dealer invoice prices and MSRPs in the United States. This in turn depends on at least two factors. First, there would have had to be, in this but-for world, a flood of significantly lower-priced Canadian cars coming across the border for resale in the United States during times of arbitrage opportunities, enough cars to cause manufacturers to take steps to protect the American market from this competition by decreasing nationally set prices. As plaintiffs themselves note, without a very large number of cars poised to cross the border, a nationwide impact on the automobile market of the sort required by plaintiffs’ theory is implausible, and the theory collapses. In our view, plaintiffs’ expert Professor Hall had not yet, at the time of class certification, fully answered such potentially relevant questions as how the size of the but-for influx of cars would be established or how large that influx would have to be to affect the national market sufficiently to raise effective dealer invoice prices and MSRPs.
Second, the plaintiffs must be able to sort out the effects of any permissible vertical restraints from the effects of the alleged, impermissible horizontal conspiracy. This question was raised below but was not fully addressed. Professor Hall asserted in a purely conclusory manner that the effects could be separated out using the concept of Nash equilibriums. If plaintiffs do not have a viable means for distinguishing between these two sets of effects, they cannot show that it was the horizontal conspiracy that caused the impact on the domestic national market upon which their theory depends.28
*28As for the second step of plaintiffs’ theory, it must include some means of determining that each member of the class was in fact injured, even if the amount of each individual injury could be determined in a separate proceeding. Predominance is not defeated by individual damages questions as long as liability is still subject to common proof. Tardiff, 365 F.3d at 6; Smi-low, 323 F.3d at 40; 6 Conte & Newberg, supra, § 18:27, at 91. This is because the class action can be limited to the question of liability, leaving damages for later individualized determinations. See Tardiff, 365 F.3d at 7; Smilow, 323 F.3d at 41; 6 Conte & Newberg, supra, § 18:53, at 179 & n. 10, § 18:56, at 190-92.29 Establishing liability, however, still requires showing that class members were injured at the consumer level. It is unclear to us how plaintiffs intend to make this connection.
The plaintiffs might intend to use their damages model to prove both fact of damages and the measure of those damages. If so, the district court would need enough information to evaluate preliminarily whether the proposed model will be able to establish, without need for individual determinations for the many millions of potential class members, which consumers were impacted by the alleged antitrust violation and which were not. See, e.g., Blades, 400 F.3d at 570 (affirming denial of class status where the actual prices paid by class members could not be determined via common proof because “[t]he amount of premiums paid, if any, is relevant to a determination of impact ... and is not merely an assessment of the amount of damages, which may be properly ascertained at a later time”); Newton, 259 F.3d at 187-88 (affirming denial of class status where plaintiffs had not only provided no model formula for measuring damages, but more fundamentally had also not demonstrated the fact of damages). “The ability to calculate the aggregate amount of damages,” as plaintiffs propose to do here, “does not absolve plaintiffs from the duty to prove each [class member] was harmed by the defendants’ practice.” Newton, 259 F.3d at 188.
The district court, while expressing skepticism regarding plaintiffs’ proposal for measuring damages, relied on this court’s consideration in Smilow of an incomplete damages model. It noted that in Smilow, this court had accepted as sufficient a representation by plaintiffs’ expert that he “could fashion” a computerized method of calculating class damages. Motor Vehicles V, 235 F.R.D. at 144 (quoting Smilow, 323 F.3d at 40) (internal quotation marks omitted). The proposed computerized model in Smilow would draw from the defendant’s records, which listed the consumers who were charged the allegedly illegal fees. Smilow, 323 F.3d at 40-41. That is, the model would have relied on data that clearly established which consumers suffered injury. Professor Hall similarly claimed that data,in defendants’ hands would provide the information he would need for his damages model; whether that data will be sufficient to establish consumer-level impact for each class member is a question that can now be answered with discovery completed.
*29Plaintiffs seem to rely on an inference that any upward pressure on national pricing would necessarily raise the prices actually paid by individual consumers. There is intuitive appeal to this theory, but intuitive appeal is not enough. Even if it is fair to assume that hard bargainers will usually pay prices closer to the dealer invoice price and poor negotiators will usually pay prices closer to the MSRP, a minimal increase in national pricing would not necessarily mean that all consumers would pay more. Too many factors play into an individual negotiation to allow an assumption — at least without further theoretical development — that any price increase or decrease will always have the same magnitude of effect on the final price paid. Even if Professor Hall’s proposed models could determine when MSRPs and dealer invoice prices were affected for which models and to what degree, it is a further question whether it can be presumed that all purchasers of those affected cars paid higher retail prices.
Some courts have allowed a presumption of class-wide impact in price-fixing cases when “the price structure in the industry is such that nationwide the conspiratorially affected prices at the wholesale level fluctuated within a range which, though different in different regions, was higher in all regions than the range which would have existed in all regions under competitive conditions.” Winoff Indus., Inc. v. Stone Container Corp. (In re Linderboard Antitrust Litig.), 305 F.3d 145, 151 (3d Cir.2002) (quoting Bogosian v. Gulf Oil Corp., 561 F.2d 434, 455 (3d Cir.1977)). If effective dealer invoice prices in the real world were equal to or greater than the effective MSRPs in the but-for world — that is, if the entire negotiating range in the but-for world would have been below the entire negotiating range in the real world — it would be easier to presume that all consumers suffered impact. The district court discussed the Bogosian presumption in its May 12, 2006 order, Motor Vehicles V, 235 F.R.D. at 138 n.35, but plaintiffs disclaim any intent to rely on the Bogosian model.
It is true that the validity of plaintiffs’ theory is a common disputed issue. Cf. Tardiff, 365 F.3d at 4-5. It will be for the fact finder to decide whether this theory is persuasive. At the class certification stage, however, the district court must still ensure that the plaintiffs’ presentation of their case will be through means amenable to the class action mechanism. We are looking here not for hard factual proof, but for a more thorough explanation of how the pivotal evidence behind plaintiffs theory can be established. If there is no realistic means of proof, many resources will be wasted setting up a trial that plaintiffs cannot win.
In sum, the district court’s oft-expressed instinct that aspects of plaintiffs’ theory remained to be developed dovetails with our own. At the time of class certification, more work remained to be done in the building of plaintiffs’ damages model and the filling out of all steps of plaintiffs’ theory of impact. Time has now passed: it is almost two years since the district court’s May 12, 2006 order, and all discovery was scheduled to be completed by March 3, 2008. The plaintiffs should now have the evidence they need to put their best foot forward, and they have had additional time to work out their models and formulas. The district court should now have a complete record before it from which to test the viability of plaintiffs’ novel theory for proving common impact.
We thus vacate and remand the certification of the state damages classes so that the district court, which has handled this case admirably so far, may reconsider those class certification orders in light of *30this opinion and the more fully developed record.
We reverse in part, vacate in part, remand in part, and order dismissal of the Clayton Act injunctive relief claim. All parties shall bear their own costs.
. The district court observed that “it is reasonable to infer that the proposed federal injunctive class numbers in the millions”; the defendants have represented to the district court that the proposed class includes "more than a hundred million American consumers.” In re New Motor Vehicles Canadian Export Antitrust Litig. (Motor Vehicles IV), No. MDL 1532, 2006 WL 623591, at *2 & n. 19 (D.Me. Mar. 10, 2006).
. Arbitrage describes the practice of simultaneously buying and selling identical securities, currency, or other assets in different markets, "with the hope of profiting from the price difference in those markets.” Black’s Law Dictionary 112 (8th ed.2004).
. A gray market is one "in which the seller uses legal but sometimes unethical methods to avoid a manufacturer’s distribution chain and thereby sell goods (esp. imported goods) at prices lower than those envisioned by the manufacturer.” Black’s Law Dictionary 989 (8th ed.2004).
. Plaintiffs distinguish between list and effective dealer invoice prices. The list dealer invoice price is determined annually for the national market by the manufacturers. Effective dealer invoice prices reflect dealer or regions-specific incentives provided by the manufacturers. Similarly, the effective MSRP takes into account direct manufacturer-to-consumer incentives, such as cash back offers. This distinction is crucial to plaintiffs' theory of antitrust impact.
. Defendants on appeal include Ford Motor Company, Ford Motor Company of Canada, Ltd., General Motors Corp., General Motors of Canada, Ltd., Honda Motor Co., Honda Canada, Inc., Nissan North America, Inc., Chrysler LLC, Chrysler Motors LLC, Chrysler Canada Inc., and Mercedes-Benz USA, LLC. In addition, the plaintiffs named two automobile dealers’ associations, the domestic National Automobile Dealers Association ("NADA”) and the Canadian Automobile Dealers Association ("CADA”). Two automakers originally named as defendants are no longer in the case: Toyota Motor Sales U.S.A., Inc. entered into a settlement agreement with plaintiffs, and plaintiffs voluntarily dismissed BMW of North America, LLC.
. The district court has not decided this issue, deeming it a merits inquiry that it would be premature to resolve at the class certification stage. Motor Vehicles V, 235 F.R.D. at 131 n. 10.
, Plaintiffs object that defendants cannot raise a standing challenge to the Rule 23(b)(2) class on appeal because they failed to present the argument to the district court. For their part, defendants claim they previously raised the issue in oral argument. Either way, courts must be vigilant about compliance with standing requirements regardless of what the parties argue, see Pagán v. Calderón, 448 F.3d 16, 26 (1st Cir.2006), and must remain so throughout the entire life cycle of a case. See generally II Areeda & Hovenkamp, Antitrust Law ¶ 335b, at 289 (2d ed.2000). The issue has not been waived.
. The court found the injunctive relief class met the requirements of Rule 23(a) — numerosity, commonality, typicality, and adequacy — then turned to the requirements of Rule 23(b)(2). Fed.R.Civ.P. 23(a); Motor Vehicles IV, 2006 WL 623591, at *2-6. Rule 23(b)(2) allows for class actions when "the party opposing the class has acted or refused to act on grounds that apply generally to the class, so that final injunctive relief or corresponding declaratory relief is appropriate respecting the class as a whole.” Fed.R.Civ.P. 23(b)(2). The court held that the defendants fall within the language of Rule 23(b)(2) because they had acted or refused to act on grounds generally applicable to the class. Motor Vehicles IV, 2006 WL 623591, at *1. The court further held that the proposed class satisfied the rule's requirement that injunctive relief be "appropriate ... with respect to the class as a whole.” Id. at *7-10.
. The district court noted the defendants’ position that because the complained-of export restrictions operated to restrict the flow of American cars into Canada as well as vice versa, consumers buying cars in the United States during periods when the value of the Canadian dollar and the prices of Canadian cars rose were actually protected from higher domestic prices. Plaintiffs' expert conceded that the defendants’ two-way restrictions could protect American consumers in such a scenario. Motor Vehicles IV, 2006 WL 623591, at *3 n. 23.
. We take judicial notice of these undisputed facts. Between 1950 and the present, it has generally cost between $0.96 and $1.40 Canadian dollar (“CAD") to purchase one United States dollar ("USD”). During that period, the exchange rate has risen significantly above $1.40 CAD to $1.00 USD only once for any length of time. This occurred between the years 1998 and 2003, when the rate ranged between $1.40 and $1.61 CAD to $1.00 USD. 5E Historical Statistics of the United States 5-572 to 5-574 (S. Carter et al. eds., 2006); Federal Reserve Statistical Release: Historical Rates for Canada, http:// www.federalreserve.gov/releases/h 10/hisl/dat 00_ca.htm. This deviation corresponds with the period during which plaintiffs’ alleged arbitrage opportunities occurred.
. The exchange rate on April 30, 2003 was $1.43 CAD to $1.00 USD, and as of publication of this opinion, the exchange rate has not once returned to that level. Instead, the relative value of the domestic dollar has been in more or less steady decline. On September 28, 2007, the value of the Canadian dollar surpassed that of the United States dollar. As of February 1, 2008, the exchange rate was $0.90 CAD to $1.00 USD. Federal Reserve Statistical Release: Historical Rates for Canada, http://www.federalreserve.gOv/releases/h 10/hist/dat00_ca.htm.
. Although the defendants challenge plaintiffs' standing to maintain the injunctive class, neither side addressed to this court any of the jurisdictional issues that might arise in the event that we vacated the federal injunctive class. We do not purport to raise an exhaustive checklist of questions to address on remand, but advert to some salient issues.
. The Class Action Fairness Act of 2005 ("CAFA"), Pub.L. 109-2, 119 Stat. at 4 (amending 28 U.S.C. § 1332) was enacted after the complaints in this action were filed. The existence of diversity jurisdiction must therefore, under the express terms of the CAFA, be determined under the pre-CAFA diversity statute. See CAFA § 9, 119 Stat. 14 ("[CAFA] shall apply to any civil action commenced on or after the date of enactment of this Act.”).
.In its September 7, 2004 order addressing supplemental jurisdiction over the state-law claims, the district court in addition asserted pendent personal jurisdiction over certain Canadian defendants based on the personal jurisdiction created by the federal antitrust claim. Motor Vehicles II, 335 F.Supp.2d at 128; see also In re New Motor Vehicles Canadian Export Antitrust Litig., 307 F.Supp.2d 145, 147-48 (D.Me.2004) (finding personal jurisdiction over certain Canadian defendants). To the extent that the district court relied on the existence of the federal claim, personal jurisdiction may yet be an issue on remand.
. The defendants did not move to dismiss all the state claims, and the plaintiffs consented to the dismissal of some of the state claims. Motor Vehicles III, 350 F.Supp.2d at 168, 175.
. The district court had requested, in a March 15, 2005 order, that each side designate three exemplar states. The plaintiffs selected Maine, Kansas, and Vermont, and the defendants chose California, New Mexico, and Tennessee.
.The Kansas class was not certified because plaintiffs conceded that the class representative had not in fact been injured by the alleged conspiracy and thus lacked standing. Motor Vehicles V, 235 F.R.D. at 131.
. To establish an antitrust claim, plaintiffs typically must prove (1) a violation of the antitrust laws, (2) an injury they suffered as a result of that violation, and (3) an estimated measure of damages. Sullivan v. Nat'l Football League, 34 F.3d 1091, 1103 (1st Cir.1994). For a class action to be appropriate, "plaintiffs need to demonstrate that common issues prevail as to [both] the existence of a conspiracy and the fact of injury.” Blades v. Monsanto Co., 400 F.3d 562, 566 (8th Cir.2005). If these two elements are established by common proof, the measure of damages can sometimes be left to individual proof, as we discuss further below.
The element of injury in the antitrust context is often referred to as "impact” or "fact of damage.” Alabama v. Blue Bird Body Co., 573 F.2d 309, 317 & n. 18 (5th Cir.1978). It is the causal link between the antitrust violation and the damages sought by plaintiffs. Sullivan, 34 F.3d at 1103. It thus requires both injury-in-fact and a showing that the injury is the result of the antitrust activity. Cordes & Co. Fin. Servs., Inc. v. A.G. Edwards & Sons, Inc., 502 F.3d 91, 106 (2d Cir.2007).
. The court reiterated in its March 21, 2007 order that it would not "make a final determination of the existence of antitrust impact at the certification stage.” Motor Vehicles VI, 241 F.R.D. at 80 (emphasis added). "Whether an illegal agreement halting Canadian imports (or removing their threat) produced antitrust causation or retail purchase price impact remains to be proven at trial or demonstrated at summary judgment.” Id. at 82 n. 5.
. Professor Kalt had described the contours of the gray market in 2000, 2001, and 2002. Professor Hall argued that the gray market in 2001 and 2002 would reflect the impact of the alleged horizontal conspiracy and that the relevant arbitrage opportunities were not present in 2000.
. For example, Professor Hall cited a study showing that dealers tend to pass on to consumers at least some pricing incentives given by manufacturers. Motor Vehicles V, 235 F.R.D. at 137 (citing M. Busse et al., $1000 Cash Back: Asymmetric Information in Auto Manufacturer Promotions 45 (Nat'l Bureau of Econ. Res. Working Paper Series No. 10887, 2004)).
. The parties filed briefs on these issues of state substantive law before the March 21, 2007 order. The court found that the parties’ briefing did not cause it to change its prior views.
. Defendants say this is not an accurate representation of California law, but do not press the point on appeal.
. Those issues were:
(a) Was there a horizontal agreement to restrict supply?
(b) Was it illegal under that particular state’s laws?
(c) ... Did the illegal agreement have antitrust or consumer protection impact in that state as the plaintiffs propose to prove it?
(d) If so, is that impact sufficient to confer standing under the particular state’s laws?
(e) How long did the conspiracy and its impact last?
Motor Vehicles V, 235 F.R.D. at 142 (footnote omitted).
. In a footnote, the district court explained that it believed it had complied with the First Circuit requirement that the district court "formulate some prediction as to how specific issues will play out,” Mowbray, 208 F.3d at 298, that it conduct a “rigorous analysis” of the Rule 23 criteria, Smilow, 323 F.3d at 38, and that it test the disputed premises "early on,” Tardiff, 365 F.3d at 4. Motor Vehicle VI, 241 F.R.D. at 81 n.7. The court noted that PolyMedica did not mandate a particular level of fact-finding by the district judge at the certification stage. Id.
. Defendants attempt to answer this problem by saying these statements demonstrate that the district court failed to inquire adequately into the record; that answer is a mismatch with the problem presented of an incomplete record and of incomplete work by the plaintiffs' expert.
. The 2003 amendments to Rule 23 deleted the provision allowing class certifications to be conditional. The advisory committee notes explain that “[a] court that is not satisfied that the requirements of Rule 23 have been met should refuse certification until they have been met.” This does not prevent a judge from modifying its certification if it becomes clear, as the case develops, that the class action vehicle is in fact inappropriate.
. While these are both questions that are themselves susceptible to common proof (the potential size of the gray market and the distinction between the effects of horizontal and vertical restraints), they go to the viability of a novel theory upon which plaintiffs rely to establish an element of their claim through common means. In that sense, these factual *28questions are akin to the question of market efficiency in securities class actions employing the fraud-on-the-market presumption of reliance. Cases like PolyMedica and Xcelera demonstrate that such factual bases of theories of common proof are appropriately, although preliminarily, tested at the class certification stage.
. The district court noted that "some states permit consumers to recover the full purchase price once liability is proven,” further simplifying the calculation of individual damages awards. Motor Vehicles V, 235 F.R.D. at 143 n. 53.