LYNCH, Chief Judge.

The Class Action Fairness Act of 2005 ("CAFA") provides for removal to federal court of state class actions that satisfy the statute's minimal diversity and class size requirements and have more than $5 million in controversy. See 28 U.S.C. §§ 1332(d), 1453. This case requires us to address for the first time the burden on a removing defendant to establish the amount in controversy under CAFA.

We hold that at least where the complaint does not contain specific damage allegations, the removing defendant must show a reasonable probability that the amount in controversy exceeds $5 million. This test uses different nomenclature from, but we believe is substantively the same as, the standards adopted by several circuits. Here, defendant failed to meet this burden, in part, because the plaintiffs' class allegations were not yet fully developed at the time of removal.

We affirm the district court's order remanding the case to state court without prejudice to the possibility that defendant may later seek removal to federal court as the state litigation progresses.

I.

In this case, the underlying dispute involves the refunding of premiums for credit insurance policies purchased in conjunction with loans to automobile buyers. Lenders who finance automobile purchases sometimes require the borrower to have life and disability insurance naming the lender as the beneficiary. This arrangement guarantees that the loan is paid back even if the borrower is injured or dies. This type of insurance is often sold as a "single premium" policy, meaning that the entire premium is paid up-front. If the borrower pays off the loan early, he may be entitled to a refund of the unearned portion of the premium — the part allocated to pay for coverage during the remaining policy period — under either the terms of the insurance contract or a state's consumer protection laws. The complaint in this case initially involved only automobile purchasers in New Hampshire, which has such a statute. See N.H.Rev.Stat. Ann. § 361-A:7(IV-a). Defendant Guarantee Trust Life Insurance Company ("GTL") sold single premium credit insurance policies to automobile purchasers in New Hampshire and forty other states through dealerships and their associated financing companies.

On June 18, 2004, Frederick Amoche and Jon Valliere sued GTL in New Hampshire state court, alleging that GTL owed them refunds of the unearned portions of their credit insurance premiums. They sought to represent a class of New Hampshire consumers who had obtained credit insurance through GTL in conjunction with an automobile loan, paid off the loan early, and had not been refunded the un-earned portion of the prepaid premium. The complaint requested money damages under a theory of breach of contract and breach of the implied covenant of good faith and fair dealing, restitution of un-earned premiums, and various forms of equitable relief, including an injunction requiring GTL to implement measures to guarantee that it would promptly refund unearned premiums to those who pay off their loans early.

On October 13, 2004, plaintiffs filed a motion for class certification as to their breach of contract claim. They defined the class as:

All persons who were sold GTL single premium credit life and/or credit disability insurance products in connection with their credit purchase of a motor vehicle from a motor vehicle dealer located within New Hampshire and who prepaid their insured loan prior to the maturity date of such credit transaction but did not receive a refund or credit for the unearned portion of the premium prior to the commencement of this law-suit. Excluded from the class are those who received payment from GTL for death or disability claims and those who are presumptive class members in a law-suit wherein the claims against GTL have been certified by a New Hampshire state court to proceed on a classwide basis.

This class definition was narrower than the one contained in the original complaint because it excluded those who already had claims pending against GTL in parallel class actions.

On December 17, 2004, plaintiffs filed a second amended complaint that conformed the class allegations to the class definition in the motion to certify and added Diane Dauphinais as a named plaintiff. On September 27, 2005, the state court certified the proposed class of New Hampshire automobile purchasers as to the plaintiffs' breach of contract claim.

Plaintiffs filed a motion for partial summary judgment on May 2, 2005, seeking to establish GTL's liability on their breach of contract claim. GTL filed a cross-motion for summary judgment on October 14, 2005. On May 2, 2006, the state court granted the plaintiffs' motion for partial summary judgment and denied GTL's cross-motion, holding that GTL's failure to refund unearned premiums where the borrower had paid off the automobile loan early violated the express terms of the insurance contract.

Having secured a finding of liability, plaintiffs filed a motion for leave to file a third amended complaint on July 12, 2007. In their proposed third amended complaint, plaintiffs expanded the class definition to include consumers from other states. As to the size of the proposed expanded class, the third amended complaint said:

Counsel has conducted an extensive preand post-filing investigation into GTL's single premium credit insurance activities across the country. Plaintiffs' counsel believe, based upon knowledge obtained in this and similar cases, that GTL has issued hundreds of thousands of credit insurance certificates pertaining to motor vehicle loans since 2000. Counsel believe, based on information obtained in this and similar cases, that a substantial percentage of the credit-insured vehicle loans were paid off early and no refunds were made.

And elsewhere, the third amended complaint alleged that GTL had harmed "thousands of class members by failing to refund their unearned premiums on early loan payoffs."

The third amended complaint also stated that "GTL has failed to refund over a million dollars in unearned premiums owed to . . . the class members"; it described the individual class members' damages as "less than $1,000" and, based upon similar suits, likely "about $200." Plaintiffs had consistently claimed that this case involved over a million dollars in unearned premium refunds. Indeed, even when the suit involved only a class of New Hampshire automobile purchasers, plaintiffs alleged that "GTL has wrongfully retained well over $1 million in unearned premiums be-longing to these Class members."

The state court held a hearing on the plaintiffs' motion for leave to amend on October 4, 2007 and granted the motion on October 15, 2007. The order granting the motion stated that "the plaintiffs seek to expand the class to include consumers from sixteen (16) other states that have enacted consumer protection legislation similar to that of New Hampshire." The order did not specify which sixteen states plaintiffs were considering, and the third amended complaint contained no geographic limitations on the proposed class. GTL received service of the court's order on October 17, 2007.

On October 23, 2007, plaintiffs filed a motion to amend the court's October 15, 2007 order, requesting that the court change the order's statement of the number of states included in the suit. Plaintiffs said:

[T]he order states that the plaintiffs are suing on behalf of 16 other states. Plaintiffs are not locked into that number, but want the Court to amend its order to merely say a range, which would be from 10 to 20 other states. This is not a substantive change, but one that would more accurately reflect a likely national scenario.

Plaintiffs did not say which ten to twenty states they were considering.

On November 15, 2007, before the state court had ruled on the plaintiffs' motion to amend, GTL filed a notice of removal in federal district court, claiming federal jurisdiction under CAFA. GTL argued that it was apparent from the face of the third amended complaint that more than $5 million was at stake in the case, pointing to its allegations that GTL had issued "hundreds of thousands" of policies, that a refund was due for "a substantial percentage" of those policies, and that the individual refund owed was "about $200." GTL argued that if the class contained more than 25,000 people — a fair reading, in its view, of "a substantial percentage" of "hundreds of thousands" — the amount in controversy would exceed $5 million.

Plaintiffs filed a motion to remand on December 5, 2007. They did not challenge the timeliness of GTL's notice of removal or argue that the suit failed to satisfy CAFA's minimal diversity and class size requirements. Instead, they argued that no federal jurisdiction existed under CAFA because less than $5 million was at stake. They claimed that the third amended complaint specified an amount of damages below the jurisdiction minimum because it stated that "GTL has failed to refund over a million dollars in unearned premiums."

Plaintiffs also directly addressed GTL's contention that the third amended complaint described a class containing more than 25,000 members. They argued that even though GTL had issued hundreds of thousands of credit insurance policies, most states would not be included within the suit. And plaintiffs noted that the proposed class included only people who had obtained loans with one of nine lenders, further limiting the class size. Because GTL had not come forward with any evidence of the potential damages in this case, plaintiffs characterized GTL's argument regarding the amount in controversy as "pure speculation" that could not support removal to federal court.

On January 9, 2008, GTL filed a memorandum in opposition to the motion to remand. In addition to restating its argument that the third amended complaint on its face established that the amount in controversy exceeded $5 million, GTL said:

[S]ince the inception of this ligation[,] . . . $452,472.29 in total unearned premium refunds have been either made or requested by or on behalf of motor vehicle purchasers who were sold GTL single premium credit life and/or credit disability insurance products in connection with the purchase of a motor vehicle from a dealer located in New Hampshire. . . .

GTL argued that if the totals from other states were similar, the amount in controversy would easily exceed $5 million in a suit involving ten to twenty additional states.

GTL attached to its memorandum an affidavit from Reed Gass, the vice president of GTL's credit insurance division, which explained how it had arrived at the $452,472.29 figure. Gass noted that GTL did not keep records of which lender the insured had used, meaning that he could not "itemize and quantify the specific number" of people who had obtained coverage from GTL through one of the nine lenders named in the third amended complaint. Nevertheless, Gass believed that the expanded class could include "numerous per-sons in all forty-one of GTL's active premium states" because several of the named lenders had significant nationwide presences.

Plaintiffs filed a reply to GTL's memorandum in opposition to the motion to remand on January 25, 2008. They challenged GTL's calculation of the relevant sum owed to people from New Hampshire, arguing that the $452,472.29 amount was inflated by including people who had not used one of the nine lenders named in the proposed class definition. Plaintiffs submitted an affidavit from Aaron Goulette, the project manager for New Hampshire Unearned Premiums Litigation, containing his calculation of the damages demanded or paid to New Hampshire consumers who fit the class definition as $221,912.03, an amount that, if representative of the dam-ages from other states, would place the amount in controversy below $5 million.

On March 25, 2008, plaintiffs filed a motion for leave to file a declaration from Kevin O'Brien, their attorney. In that declaration, O'Brien stated:

Now that GTL has provided its initial discovery responses, the plaintiffs' repeated assertions regarding the likely geographic scope of the class in this case have been proven true. Plaintiffs have informed GTL, in writing, that consistent with their representations all along, they will not seek certification of more than a 13-state class. Specifically: Connecticut, Michigan, Pennsylvania, Florida, Missouri, Illinois, New Hampshire, Texas, Indiana, New Jersey, Virginia, Maine, and Ohio.

GTL opposed the filing of the O'Brien declaration, arguing that post-removal limitations placed on the complaint could not defeat federal jurisdiction. But even if the court were to consider the O'Brien declaration, GTL noted that the size of the listed states relative to New Hampshire made it likely that the damages for the expanded class would exceed the jurisdictional minimum.

On August 14, 2008, the federal district court granted the plaintiffs' motion to remand. The court found that plaintiffs had not alleged a specific amount in damages and placed the burden on GTL to show that the amount in controversy exceeded $5 million by a preponderance of the evidence. It then addressed GTL's two theories regarding the amount in controversy: (1) that it is apparent from the face of the third amended complaint that the amount in controversy exceeds $5 million; and (2) that the calculations in the Gass affidavit show that the amount in controversy exceeds $5 million.

The district court rejected GTL's first argument, saying:

GTL's conclusion that there are over 25,000 claimants is based on conjecture because the Complaint does not allege what "a substantial percentage" of "hundreds of thousands" of credit insurance certificates is except for stating that the class includes " thousands of members" and "is composed of persons who bought GTL's credit insurance from motor vehicle dealers throughout New Hampshire and a number of other states."

Because the class size was not apparent from the face of the complaint, the court determined that the amount in controversy could not be ascertained by simple multiplication of the class size with the expected individual damages of $200.

As to GTL's second argument, the district court found that plaintiffs had "cast doubt on the $452,472.29 figure" from the Gass affidavit because that amount included those who had obtained automobile loans from lenders other than the nine specified in the class definition. It held:

The plaintiffs' theory that the total amount of damages is less than five million dollars is as persuasive as Gass's conclusion that the damages in this case are in excess of $5,400,000. Without more specific information as to the number of claimants nationwide or the amount paid by GTL in credit insurance refunds in states other than New Hampshire, the amount in controversy is uncertain.

Therefore, the court found that GTL had not carried its burden and remanded the case to state court.

On August 25, 2008, GTL petitioned for leave to appeal the district court's remand order under 28 U.S.C. § 1453(c). We granted GTL's petition on December 15, 2008. After expedited briefing, the appeal was argued on January 6, 2009.

II.

In enacting CAFA, Congress was responding to what it perceived as abusive practices by plaintiffs and their attorneys in litigating major interstate class actions in state courts, which had "harmed class members with legitimate claims and defendants that ha[d] acted responsibly," "adversely affected interstate commerce," and "undermined public respect for our judicial system." CAFA, Pub.L. No. 109-2, § 2(a), 119 Stat. 4, 4 (2005). Congress passed CAFA to prevent these harms "by providing for Federal court consideration of interstate cases of national importance under diversity jurisdiction." Id. § 2(b), 119 Stat, at 5. Beyond extending federal subject matter jurisdiction to include most major interstate class actions, CAFA also made a federal forum more accessible to removing defendants by imposing only a minimal diversity requirement, eliminating the statutory one-year time limit for removal, and providing for interlocutory appeal of a federal district court's remand order. See 28 U.S.C. §§ 1332(d)(2), 1453(b), (c).

Because CAFA is both a removal and a jurisdictional statute, we start with some basic principles from those areas. The party invoking federal jurisdiction has the burden of establishing that the court has subject matter jurisdiction over the case. See In re New Motor Vehicles Canadian Exp. Antitrust Litig., 522 F.3d 6, 14 (1st Cir. 2008). This is true generally for defendants removing to federal court. Danca v. Private Health Care Sys., Inc., 185 F.3d 1, 4 (1st Cir. 1999) ("[Removing] defendants have the burden of showing the federal court's jurisdiction.").

We now hold that burden of showing federal jurisdiction is on the defendant removing under CAFA. See, e.g., Spivey v. Vertrue, Inc., 528 F.3d 982, 986 (7th Cir. 2008) ("The removing party [under CAFA], as the proponent of federal jurisdiction, bears the burden of describing how the controversy exceeds $5 million."). This is also the conclusion reached by the seven other circuits that have considered this issue. See Strawn v. AT T Mobility LLC, 530 F.3d 293, 298 (4th Cir. 2008) (compiling cases). The parties agree that GTL, as the removing defendant, bears the burden of showing that federal jurisdiction exists. Plaintiffs and GTL, however, define that burden differently and disagree over whether GTL has carried its burden.

Before resolving these issues, there is the prior question of the applicable standard of appellate review. Plaintiffs correctly argue that our review of the district court's choice of removal standards is a legal issue subject to de novo review. But they characterize the district court's determination with respect to the amount in controversy as a factual finding, saying appellate review is for clear error. GTL, by contrast, asserts that our entire review of the district court's decision to remand is de novo. The answer is that it depends on the nature of the claimed error.

The existence of federal subject matter jurisdiction is a question of law subject to de novo review both in suits brought under CAFA, Lowery v. Ala. Power Co., 483 F.3d 1184, 1193 (11th Cir. 2007) ("We review de novo the district court's decision to remand a case to state court for lack of subject matter jurisdiction [under CAFA]."), and in other cases, Fayard v. Ne. Vehicle Servs., LLC, 533 F.3d 42, 45 (1st Cir. 2008). The ultimate question of whether jurisdiction exists is subject to de novo review. That ultimate question, however, may turn on or be influenced by the district court's role as the decider of disputed facts.

There may be removal cases in which the key facts are disputed and the district court resolves the dispute; in those situations, appellate review of that portion of the district court's assessment of subject matter jurisdiction composed of factual findings would be for clear error, cf. Skwira v. United States, 344 F.3d 64, 72 (1st Cir. 2003); Valentin v. Hosp. Bella Vista, 254 F.3d 358, 365 (1st Cir. 2001), while its ultimate assessment of jurisdiction remains subject to de novo review. Here, however, the district court made no findings of disputed fact relating to the amount in controversy, and our review is entirely de novo.

A. The Removing Defendant's Burden Regarding the Amount in Controversy Under CAFA

We agree with those circuits that have used the terminology of a "reasonable probability" as the standard a removing defendant must meet under CAFA. See Blockbuster, Inc. v. Galeno, 472 F.3d 53, 58 (2d Cir. 2006) ("[The removing defendant] must show that it appears to a 'reasonable probability' that the aggregate claims of the plaintiff class are in excess of $5 million."); Brill v. Countrywide Home Loans, Inc., 427 F.3d 446, 449 (7th Cir. 2005) ("[T]he removing litigant must show a reasonable probability that the stakes exceed the [jurisdictional] minimum.").

GTL contends that a reasonable probability burden is too rigorous and that it should only be required to show, as if it were an initial plaintiff filing, that it is not a legal certainty that the amount in controversy is less than the jurisdictional minimum. This lesser burden, GTL argues, mirrors the burden on a plaintiff who initially files in federal court. The removing defendant's effort to liken its situation to cases in which the plaintiff has chosen to be in federal court and it is the defendant who seeks to defeat federal jurisdiction does not work. In CAFA, Congress expressly expanded federal jurisdiction largely for the benefit of defendants against a background of what it considered to be abusive class action practices in state courts. See CAFA § 2, 119 Stat, at 4-5. That did not mean, however, that Congress intended to place defendants in the same position as plaintiffs who originally choose a federal forum. Congress certainly did not use any statutory language adopting the analogy, and the policy considerations are very different in the two situations.

Furthermore, placing a removing defendant in the same posture as a plaintiff who originally files in federal court would conflict with the general rule of deference to the plaintiff's chosen forum. See 14C Wright, Miller Cooper, Federal Practice and Procedure § 3725, at 95 (3d ed. 1998) (recognizing that "a greater burden [is imposed] on defendants in the removal situation than is imposed on plaintiffs who wish to litigate in federal court by invoking its original jurisdiction" to demonstrate the amount in controversy but that "[t]his discrepancy in treatment of plaintiffs and defendants may be justified by the historical tradition that the plaintiff is the master of the forum and is empowered to choose the court system and venue in which litigation will proceed").

We reject GTL's analogy. Instead, GTL, as the proponent of federal jurisdiction, must sufficiently demonstrate that the amount in controversy exceeds CAFA's jurisdictional minimum. To do so, it must show a reasonable probability that more than $5million is at stake in this case. See Blockbuster, 472 F.3d at 58; Brill, 427 F.3d at 449; cf. McNutt v. Gen. Motors Acceptance Corp. of Ind., 298 U.S. 178, 189, 56 S.Ct. 780, 80 L.Ed. 1135 (1936) ("[T]he court may demand that the party alleging jurisdiction justify his allegations by a preponderance of evidence.").

Several brief notes on this approach are in order before we apply this standard to the facts here. First, the reasonable probability standard is, to our minds, for all practical purposes identical to the preponderance standard adopted by several circuits. See Meridian Sec. Ins. Co. v. Sadowski 441 F.3d 536, 543 (7th Cir. 2006) (equating the proper application of the reasonable probability test with the preponderance standard). There was no error in the district court's discussion of GTL's burden in terms of the preponderance of the evidence. Yet because questions of removal are typically decided at the pleadings stage where little or no evidence has yet been produced, the removing defendant's burden is better framed in terms of a "reasonable probability," not a preponderance of the evidence. See Brill, 427 F.3d at 449. The "reasonable probability" language better captures the preliminary nature of this inquiry, reserving the preponderance of the evidence terminology for other conclusions.

Second, we do not wish to encourage or create a step-by-step burden shifting system, which would result in extensive and time consuming litigation over the question of the amount in controversy in CAFA removal cases. See Spielman, 251 F.3d at 4 ("[D]etermining whether a case belongs in federal court should be done quickly, without an extensive fact-finding inquiry."); Coventry Sewage Assocs. v. Dworkin Realty Co., 71 F.3d 1, 4 (1st Cir. 1995) ("[P]reliminary jurisdictional determinations should neither unduly delay, nor unfairly deprive a party from, determination of the controversy on the merits."); see also 14B Wright, Miller Cooper, supra, § 3702, at 16-17 ("The federal courts, as well as the parties, naturally are concerned about the efficient use of their own limited resources and attempt to avoid time-consuming threshold issues that do not go to the merits."). Consideration of this preliminary issue should not devolve into a mini-trial regarding the amount in controversy.

Third, with that caution in mind, deciding whether a defendant has shown a reasonable probability that the amount in controversy exceeds $5 million may well require analysis of what both parties have shown. Merely labeling the defendant's showing as "speculative" without discrediting the facts upon which it rests is insufficient. See Strawn, 530 F.3d at 299 (finding the defendant's showing sufficient to establish the amount in controversy under CAFA where the plaintiffs had "offered nothing" to challenge the accuracy of the defendant's affidavit). In the course of that evaluation, a federal court may consider which party has better access to the relevant information. See Evans v. Walter Indus., Inc., 449 F.3d 1159, 1164 n. 3 (11th Cir. 2006) ("Defendants have better access to information about conduct by the defendants, but plaintiffs have better access to information about which plaintiffs are injured and their relationship to various defendants.").

Fourth, a court's analysis of the amount in controversy focuses on whether a removing defendant has shown a reasonable probability that more than $5 million is in controversy at the time of removal. Events subsequent to removal that reduce the amount in controversy below the jurisdictional minimum do not divest a federal court of jurisdiction. Coventry Sewage Assocs., 71 F.3d at 6.

Finally, the plaintiffs' likelihood of success on the merits is largely irrelevant to the court's jurisdiction because the pertinent question is what is in controversy in the case, not how much the plaintiffs are ultimately likely to recover. See Brill, 427 F.3d at 448 ("The question is not what damages the plaintiff will recover, but what amount is 'in controversy' between the parties. That the plaintiff may fail in its proof, and the judgment be less than the threshold (indeed, a good chance that the plaintiff will fail and the judgment will be zero) does not prevent removal").

B. GTL's Ability to Show a Reasonable Probability that the Amount in Controversy Exceeds $5 Million

GTL attempts to demonstrate that the amount in controversy exceeds $5 mil-lion in two ways. First, it claims that the amount in controversy can be ascertained from the face of the third amended complaint. Second, it relies on the calculations from the Gass affidavit and the plaintiffs' statements regarding the number of states involved in the expanded class to demonstrate the amount in controversy. We decline to atomize our analysis; the entire record, as we have said, must be evaluated.

At the time that GTL removed to federal court, there was much uncertainty about which states' consumers were involved in this class action. The third amended complaint described a geographically unbounded class of automobile purchasers. But contrary to GTL's characterization of the pleadings, plaintiffs were never given permission by the state court to expand the suit to include all forty-one of GTL's active premium states. The state court granted plaintiffs leave to amend their pleadings to include only sixteen states, but those states were not identified. Plaintiffs then asked the state court to modify its order to allow them to include ten to twenty still unidentified states. GTL removed to federal court before the state court acted on the plaintiffs' motion to amend the order.

In federal court, plaintiffs have specified those states in a manner entirely consistent with their representations to the state court. They now say only thirteen states, at most, are involved, and they have identified those states. In this context, we may consider the plaintiffs' more specific and consistent statement of which states' consumers are involved in this suit. This specification is not an impermissible effort to defeat federal jurisdiction by narrowing the pleadings post-removal, see St. Paul Mercury Indem. Co. v. Red Cab Co., 303 U.S. 283, 292, 58 S.Ct. 586, 82 L.Ed. 845 (1938), but rather a fleshing out of the vague language of the third amended complaint, see Abrego Abrego, 443 F.3d at 690-91, 14C Wright, Miller Cooper, supra, § 3725, at 118 ("[A]n affidavit filed by a plaintiffs attorney subsequent to removal . . . can be considered as a clarification of an ambiguous complaint, rather than a post-removal amendment of the plaintiffs complaint. . . .").

Although the plaintiffs' still-developing class allegations made it difficult to ascertain the amount in controversy, GTL was not entirely hindered by this. It had available to it some information, at least as to the thirteen states identified. GTL was in a better position than plaintiffs to know who had purchased its policies and where they lived. Yet it provided the district court with virtually no information about the potential class size. GTL attributes this failure, in part, to its own record keeping practices. It says that it does not keep records of which lender an insured uses and is unaware of which automobile loans are paid off early until it is contacted by the lender. And in fairness to GTL, the plaintiffs' specification of which states were involved came late in the process.

These explanations only go so far. In assessing whether GTL has carried its burden of showing a reasonable probability that the amount in controversy exceeds $5 million, we may consider what information reasonably within GTL's control it failed to present in addition to any affirmative evidence of the amount in controversy. For example, a statistical analysis involving a sampling of GTL's credit insurance certificates could have given a rough sense for the class size. And GTL could have come forward with information regarding its market share and revenues from states other than New Hampshire that might have provided some insight to the court into the amount in controversy for a thirteen-state class action.

The only affirmative evidence that GTL presented of the amount in controversy is the Gass affidavit. The Gass affidavit states that the refunds requested by New Hampshire plaintiffs alone total $452,472.29. If the refunds due to plaintiffs in other states are similar, GTL argues, the amount in controversy will easily exceed $5 million in a suit involving at least eleven other states. Moreover, GTL suggests that the amount owed to plaintiffs in other states may well be higher because its insurance sales in New Hampshire represent only a small percentage of its business.

But the $452,472.29 figure for New Hampshire from the Gass affidavit cannot be translated into an estimate of the amount in controversy by simple multiplication of that sum by thirteen. That sum is not reliable. The Gass affidavit admits that the $452,472.29 amount is inflated by including those who do not fit the class definition, and plaintiffs contend that when only those who used one of the nine named lenders are counted, the sum from New Hampshire drops to $221,912.03. Moreover, the state to state differences in GTL's business practices — which could make the amount of unearned premium refunds owed in a given state either larger or smaller than the total from New Hampshire — make the amount in controversy impossible to ascertain from Gass's figures with any accuracy. Thus, the calculations from the Gass affidavit do not demonstrate a reasonable probability that the amount in controversy exceeds $5 million.

At this stage of the litigation and based on all of the information of record, at best there is a draw. GTL has not demonstrated a reasonable probability that the amount in controversy exceeds $5 million. The district court's decision to remand was proper.

III.

Our decision to affirm the district court's remand order does not permanently foreclose GTL from attempting to remove this case to federal court. Successive attempts at removal are permissible where the grounds for removal become apparent only later in the litigation. See, e.g., FDIC v. Santiago Plaza, 598 F.2d 634, 636 (1st Cir. 1979) (per curiam). Indeed, the text of the removal statute itself contemplates that a case that was not initially removable may later be removed if the basis for removal becomes apparent through a subsequent "amended pleading, motion, order or other paper." 28 U.S.C. § 1446(b).

GTL's removal of this case at the earliest possible date was understandable, given the removal statute's requirement that a defendant file a notice of removal within thirty days of his receipt of the first removable document. Id. Here, however, the litigation had not developed to a stage where GTL has shown that the requirements for federal jurisdiction were met under CAFA. Indeed, at the time of removal, plaintiffs had not yet said which states would be included within the class. It is not unfair that GTL wait until the class allegations are more fully developed before attempting to remove, if there is a later basis for removal, especially now that class actions under CAFA are exempt from the removal statute's one-year time limit. Id. § 1453(b).

The order of remand is affirmed.