In re Libor-Based Fin. Instruments Antitrust Litig.

NAOMI REICE BUCHWALD, UNITED STATES DISTRICT JUDGE

LIBOR V I I

Table of Contents

I. Introduction ...457

II. General Legal Standards ...458

1. Class Certification ...458

1.1. Standing ...458

1.2. Rule 23(a) ...459

1.3. Rule 23 (b) (3) ...461

1.4. Ascertainability ...463

1.5. Modifications to the Class Definition ...463

2. Expert Opinion ...465

2.1. The Daubert Standard ...465

2.2. Application at Class Certification ...470

III. Exchange-Based Action ...471

1. Rabobank's Daubert Motions ...473

1.1. Dr. Seyhun ...473

1.2. Dr. Netz....489

1.3. Mr. Beevers ...496

1.4. Mr. Miller ...508

2. Exchange Plaintiffs' Daubert Motions ...510

*457 2.1. Dr. Culp ...510

2.2. Dr. Hubbard ...515

2.3. Dr. Willig ...518

3. Trader-Based Manipulation Class ...528

3.1. "Fail-Safe" Class Definition ...528

3.2. Standing ...530

3.3. Ascertainability ...533

3.4. Rule 23(a) ...533

3.5. Predominance ...539

3.6. Superiority ...544

3.7. Modification of the Class Definition ...546

3.8. Conclusion ...548

4. Suppression Class ...548

4.1. Class Definition ...548

4.2. Ascertainability ...549

4.3. Rule 23(a) ...549

4.4. Predominance ...551

4.5. Superiority ...555

4.6. Conclusion ...555

IV. Lender Action ...556

1. Daubert Motions ...556

1.1. Dr. Webb ...557

1.2. Dr. Willig ...560

2. Class Certification ...563

2.1. Rule 23(a) ...563

2.2. Ascertainability ...569

2.3. Predominance ...569

2.4. Superiority ...579

2.5. Conclusion ...580

V. OTC Action ...581

1. Daubert Motion against Dr. Stiglitz ...583

2. Class Certification ...584

2.1. Standing ...584

2.2. Rule 23(a) ...585

2.3. Predominance ...590

2.4. Modification of the Class Definition ...602

2.5. Superiority ...607

2.6. Conclusion ...607

VI. Interlocutory Appealability ...608

VII. Conclusion ...609

I. INTRODUCTION

After more than six years of litigation and six substantial opinions 1 considering extensively the nature of LIBOR and its alleged manipulation, 2 we consider in this, our seventh major opinion, whether three of the cases consolidated into this multidistrict litigation should proceed as class actions: the Exchange-Based action, Metzler Investment GmbH v. Credit Suisse Group AG , No. 11 Civ. 2613; the Lender action, Berkshire Bank v. Bank of America Corp. , No. 12 Civ. 5723; and the *458 Over-the-Counter (OTC) action, Mayor of Baltimore v. Credit Suisse Group AG , No. 11 Civ. 5450. For the reasons stated below, Exchange plaintiffs' and Berkshire Bank's motions for certification of an Exchange-based class and a Lender class are denied. We grant in part and deny in part OTC plaintiffs' motion, certifying a class only as to the antitrust claims in that action.

In litigating the question of class certification, the parties have also filed ten motions to exclude expert testimony under the admissibility standards set forth in Daubert v. Merrell Dow Pharmaceuticals, Inc. , 509 U.S. 579 , 113 S.Ct. 2786 , 125 L.Ed.2d 469 (1993). The only one-line summary we can provide is that some of the Daubert motions are granted, some are denied, and others are granted in part and denied in part, as set forth in extensive detail below.

II. GENERAL LEGAL STANDARDS

We first set forth the general legal standards applicable to our resolution of the pending motions.

1. Class Certification

"The class action is 'an exception to the usual rule that litigation is conducted by and on behalf of the individual named parties only.' " Wal-Mart Stores, Inc. v. Dukes , 564 U.S. 338 , 348, 131 S.Ct. 2541 , 180 L.Ed.2d 374 (2011) (quoting Califano v. Yamasaki , 442 U.S. 682 , 700-01, 99 S.Ct. 2545 , 61 L.Ed.2d 176 (1979) ). "To come within the exception, a party seeking to maintain a class action 'must affirmatively demonstrate his compliance' with Rule 23." Comcast Corp. v. Behrend , 569 U.S. 27 , 33, 133 S.Ct. 1426 , 185 L.Ed.2d 515 (2013) (quoting Dukes , 564 U.S. at 350 , 131 S.Ct. 2541 ). When presented with a motion for class certification, we are to "assess all of the relevant evidence admitted at the class certification stage and determine whether each Rule 23 requirement has been met, just as [we] would resolve a dispute about any other threshold prerequisite for continuing a lawsuit." In re Initial Pub. Offerings Sec. Litig. (" In re IPO "), 471 F.3d 24 , 42 (2d Cir. 2006). This assessment necessarily entails the resolution of "factual disputes relevant to each Rule 23 requirement," an obligation that "is not lessened by overlap between a Rule 23 requirement and a merits issue, even a merits issue that is identical with a Rule 23 requirement." Id. at 41 ; see also Dukes , 564 U.S. at 351 , 131 S.Ct. 2541 (noting that the "rigorous analysis" of the Rule 23 requirements "will entail some overlap with the merits of the plaintiff's underlying claim"). "[T]he preponderance of the evidence standard applies to evidence proffered to establish Rule 23's requirements." Teamsters Local 445 Freight Div. Pension Fund v. Bombardier Inc. , 546 F.3d 196 , 202 (2d Cir. 2008).

1.1. Standing

Before considering the Rule 23 requirements, we first consider threshold standing issues. "Standing" in the class action context refers to two related but analytically distinct doctrines separated by a "murky line": "traditional Article III standing" on the one hand and "so-called 'class standing' " on the other. Ret. Bd. of the Policemen's Annuity & Benefit Fund v. Bank of N.Y. Mellon ("RBPA"), 775 F.3d 154 , 160 (2d Cir. 2014) ; see also LIBOR III , 27 F.Supp.3d at 480-82.

Article III standing is assessed using the oft-recited three-part formulation set forth by the Supreme Court: a "plaintiff must have (1) suffered an injury in fact, (2) that is fairly traceable to the challenged conduct of the defendant, and (3) that is likely to be redressed by a favorable judicial decision." Spokeo, Inc. v. Robins , --- U.S. ----, 136 S.Ct. 1540 , 1547, 194 L.Ed.2d 635 (2016) ; see also LIBOR III , 27 F.Supp.3d at 481 (citing *459 Lujan v. Defs. of Wildlife , 504 U.S. 555 , 560-61, 112 S.Ct. 2130 , 119 L.Ed.2d 351 (1992) ). While an injury-in-fact must be "concrete and particularized" and "actual or imminent," Lujan , 504 U.S. at 560 , 112 S.Ct. 2130 , "an injury-in-fact need not be capable of sustaining a valid cause of action," Denney v. Deutsche Bank AG , 443 F.3d 253 , 264 (2d Cir. 2006). Rather, "the fact that an injury may be outweighed by other benefits, while often sufficient to defeat a claim for damages, does not negate [Article III] standing." Id. at 265 .

In the class action context, the Second Circuit has made clear that "no class may be certified that contains members lacking Article III standing" and that any "class must therefore be defined in such a way that anyone within it would have [Article III] standing." Id. at 264 . However, when presented with a putative class, "[w]e do not require that each member ... submit evidence of personal standing." Id. at 263 . Rather, "only one of the named Plaintiffs is required to establish standing in order to seek relief on behalf of the entire class." Cent. States Se. & Sw. Areas Health & Welfare Fund v. Merck-Medco Managed Care, L.L.C. , 504 F.3d 229 , 241 (2d Cir. 2007) (emphasis added). "[P]assive members need not make any individual showing of standing, because the standing issue focuses on whether the plaintiff is properly before the court, not whether represented parties or absent class members are properly before the court." Denney , 443 F.3d at 264 (alteration in original) (quoting 1 Herbert B. Newberg & Alba Conte, Newberg on Class Actions § 2:7 (4th ed. 2002) ).

While only one named plaintiff need establish Article III standing and our analysis need not consider evidence from absent class members, we must nonetheless consider Article III standing as against each defendant. "[F]or every named defendant there must be at least one named plaintiff who can assert a claim directly against that defendant." LIBOR III , 27 F.Supp.3d at 481 (emphasis omitted) (quoting NECA-IBEW Health & Welfare Fund v. Goldman Sachs & Co. (" NECA "), 693 F.3d 145 , 159 (2d Cir. 2012) ).

"[A]t that point, [Article III] standing is satisfied and only then will the inquiry shift to a class action analysis" and the question of class standing, which refers to a named plaintiff's ability to assert claims on behalf of absent class members. NECA , 693 F.3d at 159 (quoting Merck-Medco , 504 F.3d at 241 ). "[I]n a putative class action, a plaintiff has class standing if he plausibly alleges (1) that he personally has suffered some actual injury as a result of the putatively illegal conduct of the defendant, and (2) that such conduct implicates the same set of concerns as the conduct alleged to have caused injury to other members of the putative class by the same defendants." RBPA , 775 F.3d at 161 (emphasis added) (alteration in original) (quoting NECA , 693 F.3d at 162 ). This standard, "derive[d] from constitutional standing principles" but also distinct from Article III standing itself, serves to insure that "the named plaintiff's litigation incentives are sufficiently aligned with those of the absent class members that the named plaintiff may properly assert claims on their behalf." Id. As the Second Circuit's formulation of the class standing test makes clear, class standing is assessed based on allegations rather than evidence. See, e.g. , id. (holding that named plaintiffs had satisfied the first prong of the two-prong class standing test because "they ha[d] adequately pled that they have personally suffered an actual injury as a result of [the defendant's] putatively illegal conduct").

1.2. Rule 23(a)

To proceed as a class action, each of the requirements of Rule 23(a) must be met.

*460 As relevant here, Rule 23(a) provides that "[o]ne or more members of a class may sue ... as representative parties on behalf of all members only if: (1) the class is so numerous that joinder of all members is impracticable; (2) there are questions of law or fact common to the class; (3) the claims ... of the representative parties are typical of the claims ... of the class; and (4) the representative parties will fairly and adequately protect the interests of the class." Fed. R. Civ. P. 23(a). These requirements are generally referred to as numerosity, commonality, typicality, and adequacy of representation.

Rule 23(a)(1) requires the class to be sufficiently numerous such that "joinder of all members is impracticable," Fed. R. Civ. P. 23(a)(1), but this requirement "does not mandate that joinder of all parties be impossible," Merck-Medco , 504 F.3d at 244 . Though "the numerosity inquiry is not strictly mathematical," numerosity "is presumed for classes larger than forty members." Pa. Pub. Sch. Emps.' Ret. Sys. v. Morgan Stanley & Co. , 772 F.3d 111 , 120 (2d Cir. 2014). Plaintiffs need not furnish "evidence of exact class size or identity of class members to satisfy the numerosity requirement." Robidoux v. Celani , 987 F.2d 931 , 935 (2d Cir. 1993).

The next requirement, commonality, demands that there be "questions of law or fact common to the class." Fed. R. Civ. P. 23(a)(2). A question is common to the class if it is "capable of classwide resolution-which means that determination of its truth or falsity will resolve an issue that is central to the validity of each one of the claims in one stroke." Dukes , 564 U.S. at 350 , 131 S.Ct. 2541 . That is, "[c]ommonality requires the plaintiff to demonstrate that the class members 'have suffered the same injury,' " which requires establishing more than the mere fact that class members "have all suffered a violation of the same provision of law." Id. at 349-50 , 131 S.Ct. 2541 (quoting Gen. Tel. Co. of the Sw. v. Falcon , 457 U.S. 147 , 157, 102 S.Ct. 2364 , 72 L.Ed.2d 740 (1982) ). However, "[w]here the same conduct or practice by the same defendant gives rise to the same kind of claims from all class members, there is a common question." Johnson v. Nextel Commc'ns, Inc. , 780 F.3d 128 , 137 (2d Cir. 2015) (quoting Suchanek v. Sturm Foods, Inc. , 764 F.3d 750 , 756 (7th Cir. 2014) ).

Third, "[t]o establish typicality under Rule 23(a)(3), the party seeking certification must show that 'each class member's claim arises from the same course of events and each class member makes similar legal arguments to prove the defendant's liability.' " In re Flag Telecom Holdings, Ltd. Sec. Litig. , 574 F.3d 29 , 35 (2d Cir. 2009) (quoting Robidoux , 987 F.2d at 936 ). This requirement is related to, but distinct from, the requirement of class standing. See RBPA , 775 F.3d at 161 (citing NECA , 693 F.3d at 158 n.9 ). "Typicality requires that 'the disputed issue[s] of law or fact occupy essentially the same degree of centrality to the named plaintiff's claim as to that of other members of the proposed class.' " Mazzei v. Money Store , 829 F.3d 260 , 272 (2d Cir. 2016) (internal quotation marks omitted) (quoting Caridad v. Metro-N. Commuter R.R. , 191 F.3d 283 , 293 (2d Cir. 1999), overruled on other grounds by In re IPO , 471 F.3d 24 ). Accordingly, typicality is not satisfied "where a putative class representative is subject to unique defenses which threaten to become the focus of the litigation." Baffa v. Donaldson, Lufkin & Jenrette Sec. Corp. , 222 F.3d 52 , 59 (2d Cir. 2000) (quoting Gary Plastic Packaging Corp. v. Merrill Lynch, Pierce, Fenner & Smith, Inc. , 903 F.2d 176 , 180 (2d Cir. 1990) ). "[T]he defendant need not show at the certification stage that [a] unique defense will prevail, only that it is meritorious enough to *461 require the plaintiff to devote considerable time to rebut the unique defense," In re Digital Music Antitrust Litig. , 321 F.R.D. 64 , 97 (S.D.N.Y. 2017) (quoting Lapin v. Goldman Sachs & Co. , 254 F.R.D. 168 , 179 (S.D.N.Y. 2008) ). "However, the court should not disqualify a named plaintiff based upon any groundless, far-fetched defense that the defendant manages to articulate." Lapin , 254 F.R.D. at 179 (quoting Hallet v. Li & Fung, Ltd. , No. 95 Civ. 8917 (JSM), 1997 WL 621111 , at *3 (S.D.N.Y. Oct. 6, 1997) ).

Finally, "the representative parties [must] fairly and adequately protect the interests of the class." Fed. R. Civ. P. 23(a)(4). This requirement overlaps in part with those of commonality and typicality, but adequacy of representation "also raises concerns about the competency of class counsel and conflicts of interest." Dukes , 564 U.S. at 349 n.5, 131 S.Ct. 2541 (quoting Falcon , 457 U.S. at 158 n.13, 102 S.Ct. 2364 ); see also Baffa , 222 F.3d at 60 ("Generally, adequacy of representation entails inquiry as to whether: 1) plaintiff's interests are antagonistic to the interest of other members of the class and 2) plaintiff's attorneys are qualified, experienced and able to conduct the litigation."). Accordingly, an analysis of adequacy of representation considers "whether the class representative has adequate incentive to pursue the class's claim, and whether some difference between the class representative and some class members might undermine that incentive." In re Payment Card Interchange Fee & Merch. Disc. Antitrust Litig. , 827 F.3d 223 , 231 (2d Cir. 2016). However, "[n]ot every conflict among subgroups of a class will prevent class certification-the conflict must be 'fundamental' to violate Rule 23(a)(4)." In re Literary Works in Elec. Databases Copyright Litig. , 654 F.3d 242 , 249 (2d Cir. 2011).

1.3. Rule 23(b)(3)

In addition to satisfying each requirement of Rule 23(a), a putative class must also meet "[o]ne of the bases for certification under Rule 23(b)." Roach v. T.L. Cannon Corp. , 778 F.3d 401 , 405 (2d Cir. 2015). Each class of plaintiffs here seeks to proceed under Rule 23(b)(3), which provides that a class action may proceed as such if "the court finds that the questions of law or fact common to class members predominate over any questions affecting only individual members, and that a class action is superior to other available methods for fairly and efficiently adjudicating the controversy." Fed. R. Civ. P. 23(b)(3) (emphasis added). Predominance and superiority must each be satisfied. See Sykes v. Mel S. Harris & Assocs. , 780 F.3d 70 , 82 (2d Cir. 2015) (referring to Rule 23(b)(3) as a "disjunctive inquiry"); see also Roach , 778 F.3d at 405 .

"The Rule 23(b) (3) predominance inquiry tests whether proposed classes are sufficiently cohesive to warrant adjudication by representation." Amchem Prods., Inc. v. Windsor , 521 U.S. 591 , 623, 117 S.Ct. 2231 , 138 L.Ed.2d 689 (1997). It serves to "ensure [ ] that the class will be certified only when it would 'achieve economies of time, effort, and expense, and promote ... uniformity of decision as to persons similarly situated, without sacrificing procedural fairness or bringing about other undesirable results.' " Cordes & Co. Fin. Servs., Inc. v. A.G. Edwards & Sons, Inc. , 502 F.3d 91 , 104 (2d Cir. 2007) (omission in original) (quoting Amchem , 521 U.S. at 615 , 117 S.Ct. 2231 ).

The predominance analysis entails "careful scrutiny to the relation between common and individual questions in a case." Tyson Foods, Inc. v. Bouaphakeo , --- U.S. ----, 136 S.Ct. 1036 , 1045, 194 L.Ed.2d 124 (2016). "An individual question is one where 'members of a proposed *462 class will need to present evidence that varies from member to member,' while a common question is one where 'the same evidence will suffice for each member to make a prima facie showing [or] the issue is susceptible to generalized, class-wide proof.' " Id. (alterations in original) (quoting 2 William B. Rubenstein, Newberg on Class Actions § 4:50 (5th ed. 2012) ). As its name suggests, "[t]he predominance requirement calls only for predominance, not exclusivity, of common questions." In re Visa Check/MasterMoney Antitrust Litig. , 280 F.3d 124 , 140 (2d Cir. 2001) (quoting In re Alcoholic Beverages Litig. , 95 F.R.D. 321 , 328 (E.D.N.Y. 1982) ), overruled on other grounds by In re IPO , 471 F.3d 24 . The inquiry is inherently comparative, taking into account the weight and significance of common and individual issues rather than simply their numbers. See, e.g. , In re Petrobras Sec. , 862 F.3d 250 , 268 (2d Cir. 2017) ("[P]redominance is a comparative standard.").

The existence of a single common question suffices to establish commonality, but " Rule 23(b)(3)'s predominance requirement is 'more demanding than Rule 23(a).' " Nextel , 780 F.3d at 138 (quoting Comcast , 569 U.S. at 34 , 133 S.Ct. 1426 )). Ultimately, we ask "whether issues susceptible to generalized proof outweigh individual issues," Sykes , 780 F.3d at 88 (internal quotation marks omitted) (quoting McLaughlin v. Am. Tobacco Co. , 522 F.3d 215 , 231 (2d Cir. 2008), abrogated on other grounds by Bridge v. Phx. Bond & Indem. Co. , 553 U.S. 639 , 128 S.Ct. 2131 , 170 L.Ed.2d 1012 (2008) ), or put differently, whether "common issues are 'more substantial' than individual ones," Myers v. Hertz Corp. , 624 F.3d 537 , 549 (2d Cir. 2010) (quoting Moore v. PaineWebber, Inc. , 306 F.3d 1247 , 1252 (2d Cir. 2002) ). In conducting this balancing test, we "assess (1) the 'elements of the claims and defenses to be litigated,' and (2) 'whether generalized evidence could be offered to prove those elements on a class-wide basis or whether individualized proof will be needed to establish each class member's entitlement to relief.' " Nextel , 780 F.3d at 138 (quoting 1 McLaughlin on Class Actions § 5:23 (11th ed. 2014) ); see also Erica P. John Fund, Inc. v. Halliburton Co. (" Halliburton I "), 563 U.S. 804 , 809, 131 S.Ct. 2179 , 180 L.Ed.2d 24 (2011) ("Considering whether 'questions of law or fact common to class members predominate' begins, of course, with the elements of the underlying cause of action." (quoting Fed. R. Civ. P. 23(b)(3) )). "This analysis is 'more[ ] qualitative than quantitative,' and must account for 'the nature and significance of the material common and individual issues in the case.' " In re Petrobras , 862 F.3d at 271 (alterations in original) (citations omitted) (quoting 1 William B. Rubenstein, Newberg on Class Actions § 4:50 (5th ed. 2012) ).

In addition to establishing the predominance of common questions, plaintiffs must also establish "that a class action is superior to other available methods for fairly and efficiently adjudicating the controversy." Fed. R. Civ. P. 23(b)(3). Rule 23(b)(3) sets forth four "matters pertinent" to the superiority inquiry: "(A) the class members' interests in individually controlling the prosecution or defense of separate actions; (B) the extent and nature of any litigation concerning the controversy already begun by ... class members; (C) the desirability or undesirability of concentrating the litigation of the claims in the particular forum; and (D) the likely difficulties in managing a class action." Fed. R. Civ. P. 23(b)(3). "[W]hile these factors, structurally, apply to both predominance and superiority, they more clearly implicate the superiority inquiry." Sykes , 780 F.3d at 82 ; see also In re Am. Int'l Grp., Inc. Sec. Litig. , 689 F.3d 229 , 242 (2d Cir. 2012) ("[T]he plain text of Rule 23(b)(3) states that one of the 'matters pertinent' to a finding of *463 predominance is 'the likely difficulties in managing a class action.' " (quoting Fed. R. Civ. P. 23(b)(3) )).

"While Rule 23(b)(3) sets out four individual factors for courts to consider, manageability 'is, by the far, the most critical concern in determining whether a class action is a superior means of adjudication.' " Sykes , 780 F.3d at 82 (quoting 2 William B. Rubenstein, Newberg on Class Actions § 4:72 (5th ed. 2014) ). Despite the importance of manageability, the Second Circuit has also cautioned that "failure to certify an action under Rule 23(b)(3) on the sole ground that it would be unmanageable is disfavored and should be the exception rather than the rule." In re Petrobras , 862 F.3d at 268 (quoting In re Visa Check , 280 F.3d at 140 ). Ultimately, these factors are "nonexhaustive" and nonexclusive, Amchem , 521 U.S. at 615 , 117 S.Ct. 2231 , and "assessing superiority is a fact-specific inquiry," In re Vivendi, S.A., Sec. Litig. , 838 F.3d 223 , 264 (2d Cir. 2016).

1.4. Ascertainability

In addition to the express requirements of Rule 23(a) and Rule 23 (b), the Second Circuit has "recognized an 'implied requirement of ascertainability' in Rule 23." Brecher v. Republic of Argentina , 806 F.3d 22 , 24 (2d Cir. 2015) (quoting In re IPO , 471 F.3d at 30 ). Though "the touchstone of ascertainability is whether the class is 'sufficiently definite so that it is administratively feasible for the court to determine whether a particular individual is a member,' " id. (quoting 7A Charles A. Wright et al., Federal Practice & Procedure § 1760 (3d ed. 1998) ), the Second Circuit has clarified that ascertainability does not itself require that a proposed class be administratively feasible, see In re Petrobras , 862 F.3d at 268 -69 . Rather, ascertainability imposes only a "modest threshold requirement" that asks only "whether a proposed class is defined using objective criteria that establish a membership with definite boundaries." Id. at 269 . It "will only preclude certification if a proposed class definition is indeterminate in some fundamental way." Id.

Given that ascertainability is "distinct from the predominance requirement," id. at 264 n.15 (quoting In re IPO , 471 F.3d at 45 ), we do not read In re Petrobras to preclude a consideration of administrative feasibility concerns in analyzing predominance and superiority. Indeed, In re Petrobras reasoned that a freestanding administrative feasibility requirement as part of ascertainability would be duplicative of the manageability factor of the superiority inquiry (if administrative feasibility were considered comparatively rather than absolutely), id. at 268, and would "risk[ ] encroaching on territory belonging to the predominance requirement, such as classes that require highly individualized determinations of member eligibility," id. (citing Mazzei , 829 F.3d at 272 ). Accordingly, to the extent that any of the putative classes present administrative feasibility concerns, we will consider those issues not as part of the ascertainability analysis, but as part of the Rule 23(b)(3) predominance and superiority inquiries.

1.5. Modifications to the Class Definition

As we have recognized in denying defendants' motions to strike plaintiffs' class-action allegations, "[a] court is not bound by the class definition proposed in the complaint and should not dismiss the action simply because the complaint seeks to define the class too broadly." May 13, 2016 Order, 2016 WL 2851333 , at *2, slip op. at *3 (quoting Robidoux , 987 F.2d at 937 ), ECF No. 1408. Rule 23(c)(4) and Rule 23(c)(5) provide two specific means of modification, but a court also "has broad *464 discretion to modify the class definition as appropriate." 5 Moore 's Federal Practice § 23.21[6] (3d ed. 2017).

Rule 23(c)(4) provides that "[w]hen appropriate, an action may be brought or maintained as a class action with respect to particular issues." Fed. R. Civ. P. 23(c)(4). The rule may be applied "to certify a class on a particular issue even if the action as a whole does not satisfy Rule 23(b)(3)'s predominance requirement." In re Nassau Cty. Strip Search Cases , 461 F.3d 219 , 225 (2d Cir. 2006). Additionally, "[i]f an action includes multiple claims, one or more of which might qualify as a certifiable class claim, the court may separate such claims from other claims in the action and certify them under the provisions of subsection (c)(4) of Rule 23." 1 McLaughlin on Class Actions § 4:44 (14th ed.) (Westlaw 2017) (internal quotation marks omitted); see also, e.g. , In re AMF Bowling Sec. Litig. , No. 99 Civ. 3023 (DC), 2002 WL 461513 (S.D.N.Y. Mar. 26, 2002).

The Second Circuit has instructed that "[d]istrict courts should take full advantage of this provision to certify separate issues," Robinson v. Metro-N. Commuter R.R. Co. , 267 F.3d 147 , 167 (2d Cir. 2001) (alterations incorporated) (internal quotation marks omitted), abrogated on other grounds by Dukes , 564 U.S. 338 , 131 S.Ct. 2541 , but has also recognized that issue certification may be inappropriate if a "number of questions ... would remain for individual adjudication" or if issue certification "would not materially advance the litigation because it would not dispose of larger issues" relevant to the case. McLaughlin , 522 F.3d at 234 .

That is, "the rule should not be invoked merely to postpone confronting difficult certification questions," 7AA Charles A. Wright et al., Federal Practice & Procedure § 1790 (3d ed.) (Westlaw 2017), and indeed, overly aggressive application of Rule 23(c)(4) would nullify Rule 23(b)(3)'s predominance requirement, as a class may be certified only as to the common issues raised, cf. Castano v. Am. Tobacco Co. , 84 F.3d 734 , 745 n.21 (5th Cir. 1996) ("[T]he result would be automatic certification in every case where there is a common issue, a result that could not have been intended."). "[A] class action movant cannot gerrymander predominance by suggesting that only a single issue be certified for class treatment (in which, by definition, it will 'predominate')" when more substantial individual issues remain. 1 McLaughlin on Class Actions § 4:43 (14th ed.) (Westlaw 2017) (quoting Hyderi v. Wash. Mut. Bank, FA , 235 F.R.D. 390 , 398 (N.D. Ill. 2006) ).

Paralleling Rule 23(c)(4), Rule 23(c)(5) provides that "[w]hen appropriate, a class may be divided into subclasses that are each treated as a class under this rule." Fed. R. Civ. P. 23(c)(5). 3 For example, when conflicts exist between class members, they "can be cured by dividing the class into separate 'homogeneous subclasses ... with separate representation to eliminate conflicting interests.' " In re Literary Works , 654 F.3d at 249 -50 (quoting Ortiz v. Fibreboard Corp. , 527 U.S. 815 , 856, 119 S.Ct. 2295 , 144 L.Ed.2d 715 (1999) ). However, the placement of plaintiffs into "multiple subclasses *465 ... can generate unnecessary administrative inefficiencies," 3 William B. Rubenstein, Newberg on Class Actions § 7:30 (5th ed.) (Westlaw 2017) (footnote omitted), and "at some point there must be an end to reclassification," Ortiz , 527 U.S. at 857 , 119 S.Ct. 2295 ; see also In re Literary Works , 654 F.3d at 257 (describing seven subclasses as "surely beyond the point at which 'reclassification with separate counsel' must end"). "The necessity of a large number of subclasses may indicate that common questions do not predominate," and "[t]he creation of a number of subclasses ... may make the case unmanageable [and] may defeat the superiority requirement." Manual for Complex Litigation § 21.23 (4th ed. 2004).

Rule 23(c)(4) and Rule 23(c)(5) are phrased permissively. Consistent with the text of the rule, "[t]he court ... is not obligated to implement Rule 23(c)(4) [and Rule 23(c)(5) ] on its own initiative" by certifying classes only as to certain issues or creating subclasses. Lundquist v. Sec. Pac. Auto. Fin. Servs. Corp. , 993 F.2d 11 , 14 (2d Cir. 1993) (citing U.S. Parole Comm'n v. Geraghty , 445 U.S. 388 , 408, 100 S.Ct. 1202 , 63 L.Ed.2d 479 (1980) ); see also 7AA Charles A. Wright et al., Federal Practice & Procedure § 1790 (3d ed.) (Westlaw 2017) ("[T]he trial court has no independent obligation to utilize Rule 23(c)(4) sua sponte."). It remains "plaintiff's burden to show how the action may be [modified] to avoid certification problems." Lundquist , 993 F.2d at 14 (quoting Geraghty , 445 U.S. at 408 , 100 S.Ct. 1202 ). 4 Similarly, the decision to otherwise modify a class definition is a discretionary one. See Mazzei v. Money Store , 288 F.R.D. 45 , 55 (S.D.N.Y. 2012) ("[C]ourts have the discretion to construe the complaint or redefine the class to bring it within the scope of Rule 23." (internal quotation marks omitted)); see also 7A Charles A. Wright et al., Federal Practice & Procedure § 1759 (3d ed.) (Westlaw 2017).

Accordingly, in order to avoid "the perverse effect of turning defense counsel and the Court into plaintiffs' counsel's co-counsel, with plaintiffs waiting to see what objections defendants raise and how the Court rules on those objections and then amending their [proposed class definitions] as necessary based on what they learned in the process," LIBOR II , 962 F.Supp.2d at 626-27, we will consider issue certification under Rule 23(c)(4), subclass creation under Rule 23(c)(5), and other modifications of the proposed class definitions under our discretionary authority only where plaintiffs have set forth such proposals in sufficient detail.

2. Expert Opinion

2.1. The Daubert Standard

Expert testimony is admissible under Rule 702 of the Federal Rules of Evidence, which provides in full:

A witness who is qualified as an expert by knowledge, skill, experience, training, or education may testify in the form of an opinion or otherwise if: (a) the expert's scientific, technical, or other specialized knowledge will help the trier of fact to understand the evidence or to determine a fact in issue; (b) the testimony is based on sufficient facts or data; (c) the testimony is the product of reliable principles and methods; and (d) the *466 expert has reliably applied the principles and methods to the facts of the case.

Under the Supreme Court's decisions in Daubert v. Merrell Dow Pharmaceuticals, Inc. , 509 U.S. 579 , 113 S.Ct. 2786 , 125 L.Ed.2d 469 (1993), and Kumho Tire Co. v. Carmichael , 526 U.S. 137 , 119 S.Ct. 1167 , 143 L.Ed.2d 238 (1999), we have a " 'gatekeeping' function under Rule 702," under which we are "charged with 'the task of ensuring that an expert's testimony both rests on a reliable foundation and is relevant to the task at hand.' " Amorgianos v. Nat'l R.R. Passenger Corp. , 303 F.3d 256 , 265 (2d Cir. 2002) (quoting Daubert , 509 U.S. at 597 , 113 S.Ct. 2786 ). "[T]he proponent of expert testimony has the burden of establishing by a preponderance of the evidence that the admissibility requirements of Rule 702 are satisfied." United States v. Williams , 506 F.3d 151 , 160 (2d Cir. 2007).

The Second Circuit has distilled Rule 702's requirements into three broad criteria: (1) qualifications, (2) reliability, and (3) relevance and assistance to the trier of fact. See Nimely v. City of New York , 414 F.3d 381 , 396-97 (2d Cir. 2005).

2.1.1. Qualifications

When presented with expert testimony, we first consider "the threshold question of whether a witness is 'qualified as an expert by knowledge, skill, experience, training, or education' to render his or her opinions." Id. at 396 n.11 (quoting Fed. R. Evid. 702 ). We analyze "the totality of the witness's background to determine whether he or she exhibits any one or more of the qualifications listed in Rule 702... with respect to a relevant field." Washington v. Kellwood Co. , 105 F.Supp.3d 293 , 304 (S.D.N.Y. 2015).

As Rule 702's use of the disjunctive suggests, "any one of [the] five forms of qualifications will satisfy the rule." Tiffany (NJ) Inc. v. eBay, Inc. , 576 F.Supp.2d 457 , 458 (S.D.N.Y. 2007). "A formal education in a particular field is sufficient to qualify a witness as an expert" as a general matter, such that a "lack of extensive practical experience directly on point does not necessarily preclude [the] expert from testifying." Cary Oil Co. v. MG Ref. & Mktg., Inc. , No. 99 Civ. 1725 (VM), 2003 WL 1878246 , at *2 (S.D.N.Y. Apr. 11, 2003) (internal quotation marks omitted). Similarly, "a lack of formal training does not necessarily disqualify an expert from testifying if he or she has equivalent relevant practical experience." In re Rezulin Prods. Liab. Litig. , 309 F.Supp.2d 531 , 559 (S.D.N.Y. 2004).

Indeed, "[c]ourts in this circuit have noted that an expert should not be required to satisfy an overly narrow test of his own qualifications." United States v. Tuzman , No. 15 Cr. 536 (PGG), 2017 WL 6527261 , at *9 (S.D.N.Y. Dec. 18, 2017) (collecting cases) (quoting Arista Records LLC v. Lime Grp. LLC , No. 06 Civ. 5936 (KMW), 2011 WL 1674796 , at *3 (S.D.N.Y. May 2, 2011) ). "[T]he Second Circuit [has] allowed an expert to testify as to matters within his general expertise even though he lacked qualifications as to certain technical matters within that field." Pension Comm. of the Univ. of Montreal Pension Plan v. Banc of Am. Sec., LLC , 691 F.Supp.2d 448 , 457 (S.D.N.Y. 2010) (citing McCullock v. H.B. Fuller Co. , 61 F.3d 1038 , 1042-43 (2d Cir. 1995) ).

We then "compare the area in which the witness has superior knowledge, education, experience, or skill with the subject matter of the proffered testimony," which must overlap. United States v. Tin Yat Chin , 371 F.3d 31 , 40 (2d Cir. 2004). "[I]t is worth emphasizing that, because a witness qualifies as an expert with respect to certain matters or areas of knowledge, it by no means follows that he or she is qualified to express expert opinions as to other fields." Nimely , 414 F.3d at 399 n.13.

*467 Conversely, an expert's lack of qualifications as to some of the opinions offered does not render inadmissible the opinions that he is qualified to offer.

2.1.2. Reliability

We next determine "whether the proffered testimony has a sufficiently 'reliable foundation' to permit it to be considered." Campbell ex rel. Campbell v. Metro. Prop. & Cas. Ins. Co. , 239 F.3d 179 , 184 (2d Cir. 2001) (quoting Daubert , 509 U.S. at 597 , 113 S.Ct. 2786 ). "In this inquiry, [we] should consider the indicia of reliability identified in Rule 702, namely, (1) that the testimony is grounded on sufficient facts or data; (2) that the testimony 'is the product of reliable principles and methods'; and (3) that 'the witness has applied the principles and methods reliably to the facts of the case.' " Amorgianos , 303 F.3d at 265 (quoting Fed. R. Evid. 702).

The Supreme Court has identified a number of factors that may be considered in assessing reliability: "(1) whether a theory or technique 'can be (and has been) tested,' (2) 'whether the theory or technique has been subjected to peer review and publication,' (3) a technique's 'known or potential rate of error,' and 'the existence and maintenance of standards controlling the technique's operation,' and (4) whether a particular technique or theory has gained 'general acceptance' in the relevant scientific community." Id. at 266 (citations omitted) (quoting Daubert , 509 U.S. at 593 -94 , 113 S.Ct. 2786 ). These factors are not a "definitive checklist or test," Daubert , 509 U.S. at 593 , 113 S.Ct. 2786 , as "the gatekeeping inquiry must be tied to the facts of a particular case," Kumho Tire , 526 U.S. at 150 , 119 S.Ct. 1167 (internal quotation marks omitted), and "will necessarily vary from case to case," Amorgianos , 303 F.3d at 266 . "[T]he law grants a district court the same broad latitude when it decides how to determine reliability as it enjoys in respect to its ultimate reliability determination." Restivo v. Hessemann , 846 F.3d 547 , 576 (2d Cir. 2017) (emphasis omitted) (quoting Kumho Tire , 526 U.S. at 142 , 119 S.Ct. 1167 ). 5 Accordingly, "[i]n assessing the reliability of an expert opinion, a resort to common sense is not inappropriate." Johnson Elec. N. Am. Inc. v. Mabuchi Motor Am. Corp. , 103 F.Supp.2d 268 , 286 (S.D.N.Y. 2000).

Though "flexible," the reliability inquiry "must focus on the principles and methodology employed by the expert, without regard to the conclusions the expert has reached or [our] belief as to the correctness of those conclusions." Amorgianos , 303 F.3d at 266 . The expert's methodology is to be assessed step-by-step, and "it is critical that an expert's analysis be reliable at every step." Id. at 267 . " [A]ny step that renders the analysis unreliable under the Daubert factors renders the expert's testimony inadmissible." Id. (emphasis omitted) (quoting In re Paoli R.R. Yard PCB Litig. , 35 F.3d 717 , 745 (3d Cir. 1994) ).

"But conclusions and methodology are not entirely distinct from one another." Gen. Elec. Co. v. Joiner , 522 U.S. 136 , 146, 118 S.Ct. 512 , 139 L.Ed.2d 508 (1997). "[N]othing in either Daubert or the Federal Rules of Evidence requires a district *468 court to admit opinion evidence that is connected to existing data only by the ipse dixit of the expert," and "[a] court may conclude that there is simply too great an analytical gap between the data and the opinion proffered." Id. That is, "when an expert opinion is based on data, a methodology, or studies that are simply inadequate to support the conclusions reached, Daubert and Rule 702 mandate the exclusion of that unreliable opinion testimony." Amorgianos , 303 F.3d at 266 .

We offer three additional observations regarding the incomplete distinction between "methodology" on the one hand and "conclusions" and "results" on the other. First, a challenge to an expert's methodology will necessarily call into question the conclusions derived from the application of that methodology. Such a challenge does not impermissibly attack an expert's results simply because those results are collaterally damaged by the challenge directed towards the expert's methodology.

Second, robustness testing of an expert's methodology-by applying that methodology to different data or with different assumptions and examining the results produced by the methodology so applied-is not an impermissible challenge to the expert's results. Rather, this robustness and sensitivity testing relates directly to two of the Daubert factors articulated by the Supreme Court: whether the methodology "can be (and has been) tested" and the methodology's "known or potential rate of error." Daubert , 509 U.S. at 594 , 113 S.Ct. 2786 . Robustness testing and sensitivity testing that produces contradictory or otherwise implausible results strongly suggest that a methodology has been insufficiently tested and that the methodology has a high potential rate of error.

Third, inconsistent results are an "indicia of unreliability" in an expert's methodologies. Lippe v. Bairnco Corp. , 99 Fed.Appx. 274 , 279 (2d Cir. 2004). This principle is clearest in the context of inconsistent results produced by the same methodology. See, e.g. , Louis Vuitton Malletier v. Dooney & Bourke, Inc. , 525 F.Supp.2d 558 , 569 (S.D.N.Y. 2007) (finding "unexplained inconsistency between the results" produced by two iterations of the same methodology to be a basis for exclusion); United States v. Hermanek , 289 F.3d 1076 , 1097 (9th Cir. 2002) ("Inconsistent results may be an indicator of unreliability."). However, it is no less applicable to multiple methodologies intended to measure the same phenomenon that ultimately produce inexplicably inconsistent results. See Lippe , 99 Fed.Appx. at 279 ; see also In re Paoli , 35 F.3d at 777 (identifying "inconsistent results" produced by two analyses as supporting the exclusion of an expert's evidence).

"When faced with expert testimony that contains both reliable and unreliable opinions, district courts often exclude only the unreliable testimony." In re Pfizer Inc. Sec. Litig. , 819 F.3d 642 , 665 (2d Cir. 2016). If "the unreliable portion of an opinion can easily be distinguished from testimony that could help the [trier of fact], it may be an abuse of discretion to throw the good out with the bad." Id. However, "[we] are 'not obligated to prune away all of the problematic' elements of an expert's proposed testimony 'to save the remaining portions, however small.' " Id. (quoting Bricklayers & Trowel Trades Int'l Pension Fund v. Credit Suisse Sec. (USA) LLC , 752 F.3d 82 , 96 (1st Cir. 2014) ).

2.1.3. Relevance and Assistance to the Factfinder

"Even after determining that a witness is 'qualified as an expert' to testify as to a particular matter, and that the opinion is based upon reliable data and methodology, Rule 702 [ (a) ] requires the district court to make a third inquiry: whether the expert's testimony ... will *469 'assist the trier of fact' " in understanding the evidence or determining a fact in issue. Nimely , 414 F.3d at 397 (citations omitted) (quoting Fed. R. Evid 702 ). "This condition goes primarily to relevance," Daubert , 509 U.S. at 591 , 113 S.Ct. 2786 , as "[e]xpert testimony which does not relate to any issue in the case is not relevant and, ergo, non-helpful," Raskin v. Wyatt Co. , 125 F.3d 55 , 66 (2d Cir. 1997) (alteration in original) (quoting Daubert , 509 U.S. at 591 , 113 S.Ct. 2786 ). Relevance, in turn, is assessed with respect to Rule 401 of the Federal Rules of Evidence : "whether it 'ha[s] any tendency to make the existence of any fact that is of consequence to the determination of the action more probable or less probable than it would be without the evidence.' " Campbell , 239 F.3d at 184 (alteration in original) (quoting Fed. R. Evid. 401 ).

Accordingly, the Second Circuit has "consistently held ... that expert testimony that usurps either the role of the trial judge" in determining "the applicable law or the role of the [trier of fact] in applying that law to the facts before it" is inadmissible because it "by definition does not aid the [trier of fact] in making a decision," Nimely , 414 F.3d at 397 (quoting United States v. Bilzerian , 926 F.2d 1285 , 1294 (2d Cir. 1991) and United States v. Duncan , 42 F.3d 97 , 101 (2d Cir. 1994) ) (alterations incorporated and citations omitted).

Similarly, "expert testimony that seeks to address 'lay matters which [the trier of fact] is capable of understanding and deciding without the expert's help' is not relevant and is therefore inadmissible." United States v. Jiau , 734 F.3d 147 , 154 (2d Cir. 2013) ( quoting Andrews v. Metro N. Commuter R.R. Co. , 882 F.2d 705 , 708 (2d Cir. 1989) ). That is, testimony addressing lay matters is not based on an "expert's scientific, technical, or other specialized knowledge," and therefore fails to satisfy the first part of Rule 702(a). Accordingly, it is "inappropriate for experts to become a vehicle for factual narrative." SEC v. Tourre , 950 F.Supp.2d 666 , 675 (S.D.N.Y. 2013) (citing, inter alia , Highland Capital Mgmt., L.P. v. Schneider , 551 F.Supp.2d 173 , 187 (S.D.N.Y. 2008) ). 6

As with qualifications and reliability, we also disaggregate an expert's opinions before assessing their relevance and helpfulness: the fact that some of an expert's opinions are irrelevant does not render all of the expert's opinions inadmissible. Nonetheless, we need not overly fragment an expert's opinions in order to pick out only the relevant and helpful portions. Cf. In re Pfizer , 819 F.3d at 665 . 7

*470 2.2. Application at Class Certification

"Neither the Supreme Court nor the Second Circuit has definitely decided whether the Daubert standard governs the admissibility of expert evidence submitted at the class certification stage." Adkins v. Morgan Stanley , 307 F.R.D. 119 , 148 (S.D.N.Y. 2015) (quoting Chen-Oster v. Goldman, Sachs & Co. , 114 F.Supp.3d 110 , 114 (S.D.N.Y. 2015) ). Indeed, the issue remains unsettled nationally and in this district. Compare, e.g. , In re Blood Reagents Antitrust Litig. , 783 F.3d 183 , 187 (3d Cir. 2015) ("[A] plaintiff cannot rely on challenged expert testimony, when critical to class certification, to demonstrate conformity with Rule 23 unless the plaintiff also demonstrates, and the trial court finds, that the expert testimony satisfies the standard set out in Daubert ."); Messner v. Northshore Univ. HealthSystem , 669 F.3d 802 , 812-14 (7th Cir. 2012) ; Ellis v. Costco Wholesale Corp. , 657 F.3d 970 , 982 (9th Cir. 2011) ; Sher v. Raytheon Co. , 419 Fed.Appx. 887 , 890-91 (11th Cir. 2011) ; and Ge Dandong v. Pinnacle Performance Ltd. , No. 10 Civ. 8086 (JMF), 2013 WL 5658790 , at *13 (S.D.N.Y. Oct. 17, 2013), with, e.g. , In re Zurn Pex Plumbing Prods. Liab. Litig. , 644 F.3d 604 , 613 (8th Cir. 2011) ("The main purpose of Daubert exclusion is to protect juries from being swayed by dubious scientific testimony. That interest is not implicated at the class certification stage where the judge is the decision maker."), and In re Scotts EZ Seed Litig., 304 F.R.D. 397 , 412 n.8 (S.D.N.Y. 2015) (analyzing "whether each of [the expert's] proposed methodologies satisfy Comcast " rather than conducting a Daubert analysis).

We are persuaded by the view that expert evidence submitted at the class certification stage is subject to the Daubert standard. First, the Supreme Court has suggested in dicta that Daubert applies, commenting that "the District Court concluded that Daubert did not apply to expert testimony at the certification stage of class-action proceedings. We doubt that is so...." Dukes , 564 U.S. at 354 , 131 S.Ct. 2541 . Not only has the Second Circuit characterized this statement as "suggesting that a Daubert analysis may be required at least in some circumstances," In re U.S. Foodservice Inc. Pricing Litig. , 729 F.3d 108 , 129 (2d Cir. 2013), but this statement (though dicta) is also "the only discussion of [the issue] by the Supreme Court of which we are aware" such that we consider it "to be persuasive authority here," Tiffany (NJ) Inc. v. eBay Inc. , 600 F.3d 93 , 108 (2d Cir. 2010).

Second, we interpret the Second Circuit's decisions in In re IPO , Bombardier , and In re U.S. Foodservice , as supporting a more searching examination of expert testimony offered at the class certification stage. See 1 McLaughlin on Class Actions § 3:14 (14th ed.) (Westlaw 2017) (describing the Second Circuit's "substantial expansion of the extent to which district courts may evaluate expert testimony on class certification" and citing, inter alia, In re IPO and Bombardier ). Though the Second Circuit did not decide in In re U.S. Foodservice whether or when a Daubert analysis forms a necessary component of a district court's rigorous analysis," it reasoned that "[i]n In re IPO , we disavowed our earlier statement that 'an expert's testimony may establish a component of a Rule 23 requirement simply by not being fatally flawed.' " In re U.S. Foodservice , 729 F.3d at 129 -30 (quoting In re IPO , 471 F.3d at 41 ). Indeed, given the Second Circuit's direction that we consider "the relevant evidence admitted at the class certification stage," In re IPO , 471 F.3d at 42 (emphasis added), it appears to us that the standard rules of evidence should apply, see Chen-Oster , 114 F.Supp.3d at 114 (" Daubert is an amplification of *471 Rule 702, and the Federal Rules of Evidence apply generally to 'proceedings' in the courts of the United States.").

Third, at least two leading treatises endorse this view. "An undiluted Daubert analysis is consonant with a class certification standard that requires a determination by a preponderance of the evidence that each Rule 23 requirement has been met, and under which head-to-head weighing of competing expert evidence is proper." 1 McLaughlin on Class Actions § 3:14 (14th ed.) (Westlaw 2017). "It is not sufficient for the court simply to determine that the testimony could evolve into something admissible by the time of trial." 7AA Charles A. Wright et al., Federal Practice & Procedure § 1785 (4th ed.) (Westlaw 2017).

Though we conclude that Daubert applies, our inquiry is guided by the purpose for which the evidence is introduced-establishing the various class certification requirements. "[T]he question is not whether a jury at trial should be permitted to rely on the expert's report to find facts as to liability, but rather whether the Court may utilize it in deciding whether the requisites of Rule 23 have been met." Pinnacle Performance , 2013 WL 5658790 , at *13 (alterations incorporated) (quoting In re Visa Check/MasterMoney Antitrust Litig. , 192 F.R.D. 68 , 77 (E.D.N.Y. 2000) ); see also Fort Worth Emps.' Ret. Fund v. J.P. Morgan Chase & Co. , 301 F.R.D. 116 , 126 (S.D.N.Y. 2014) (limiting the Daubert inquiry "to whether or not the expert reports are admissible to establish the requirements of Rule 23" (quoting Pinnacle Performance , 2013 WL 5658790 , at *13 )).

However, "[a] conclusion that proffered expert evidence is sufficiently reliable and relevant to pass Daubert muster does not end the inquiry on class certification." 1 McLaughlin on Class Actions § 3:14 (14th ed.) (Westlaw 2017); see In re IPO , 471 F.3d at 42 ("[W]e also disavow the suggestion ... that an expert's testimony may establish a component of a Rule 23 requirement simply by being not fatally flawed."). Rather, expert opinion deemed to be admissible comprises only one part of "the relevant evidence admitted at the class certification stage" to be weighed in determining "whether each Rule 23 requirement has been met," In re IPO , 471 F.3d at 42 , and each requirement must still be established by a preponderance of the evidence, see Bombardier , 546 F.3d at 202 .

Therefore, to the extent that flaws in expert testimony proffered at class certification do not warrant that testimony's exclusion by the Court as gatekeeper under Daubert at the threshold, those flaws may nonetheless be considered in the Rule 23 analysis undertaken by the Court as trier of fact. Put differently, though a Daubert motion is an improper venue in which to take sides in a "battle of the experts" offered by competing parties, In re Joint E. & S. Dist. Asbestos Litig. , 52 F.3d 1124 , 1135 (2d Cir. 1995), disputes between experts must be resolved if necessary to the Rule 23 analysis, see In re IPO , 471 F.3d at 42 .

III. EXCHANGE-BASED ACTION

Exchange plaintiffs seek certification of a class comprised of traders of Eurodollar futures (EDF) contracts and options on EDF contracts. Specifically, they seek certification of a class defined as follows:

All persons, corporations and other legal entities (other than Defendants, their employees, affiliates, parents, subsidiaries, and co-conspirators) ("Eligible Persons") that transacted in Eurodollar futures and options on Eurodollar futures on the Chicago Mercantile Exchange between January 1, 2005 and May 17, 2010 (the "Class Period") and that were harmed or satisfy one or more of "A," "B," or "C" below. 8

*472 The proposed class definition then defines three subcriteria, the second of which consists of three sub-subcriteria and the third of which consists of two sub-subcriteria:

SUBPART A. Eligible Persons that sold a Eurodollar futures contract, or bought a put option or sold a call option on Eurodollar futures before August 7, 2007 and purchased all or part of this short position back at the final expiration formula price of a Eurodollar futures contract expiring after August 7, 2007 and before May 17, 2010.

SUBPART B. Eligible Persons that (1) purchased Eurodollar futures contract(s) or call options on Eurodollar futures on the following dates: April 7, 2006, August 17, 2006, October 26, 2006, and December 22, 2006; or (2) sold Eurodollar futures contracts or purchased put options on Eurodollar futures on the following dates: September 29, 2005, November 28, 2005, June 30, 2006, September 1, 2006, November 29, 2006, February 28, 2007, March 1, 2007, July 30, 2007, and August 6, 2007; or (3) purchased or sold Eurodollar futures contracts (or options) and that were harmed between January 1, 2005 and August 6, 2007 inclusive. 9

SUBPART C. Eligible Persons that initiated a Eurodollar futures contract or options position on or after April 15, 2009 and on or before May 17, 2010 ("Period 3"), and who satisfy "1" or "2" below.

1. Eligible Persons included in "C" are those that purchased or sold a Eurodollar futures or options contract to initiate a position during Period 3 and that were harmed.

2. Eligible Persons included in "C" are also those that purchased a Eurodollar futures contract (including Eurodollar futures contracts the expiration for which was less than 365 calendar days after the date of such purchase) to initiate a long position during Period 3, and continued to hold all or part of such long position until liquidating the position after Period 3.

The operative Corrected Fourth Amended Consolidated Class Action Complaint, identified panel banks Bank of America, Barclays, Citi, Deutsche Bank, JPMorgan Chase, Rabobank, and UBS (and certain affiliates of these panel banks) as defendants. (Corrected Fourth Am. Consolidated Class Action Compl. ("Corrected 4AC") ¶¶ 34-38, 45, 57-59, 66-68, 77-78, Dec. 11, 2017, ECF No. 2363.) Exchange plaintiffs assert five claims, four under the Commodities Exchange Act (CEA) and the fifth under the Sherman Act. The CEA claims are asserted against all defendants, whereas the antitrust claims are asserted against only Bank of America, Citi, and JPMorgan Chase consistent with our rulings in LIBOR VI , see 2016 WL 7378980 , at *25, slip op. at app. A-1. (Corrected 4AC ¶¶ 668-705.)

The scope of claims remaining in this action has also been narrowed by several settlements between Exchange plaintiffs and particular defendants. We preliminarily approved a settlement with Barclays, *473 see Dec. 2, 2014 Order, 2014 WL 6851096 , ECF No. 861, and we deferred preliminary approval of the settlements with Citi and Deutsche Bank (Letter from Christopher Lovell & David Kovel to the Court, Oct. 11, 2017, ECF No. 2307) pending our resolution of the class-certification motions. Bank of America and JPMorgan Chase have also reached settlements with Exchange plaintiffs, though no documentation memorializing the settlement has yet been filed. 10 (Jan. 18, 2018 Hr'g Tr. ("Hr'g Tr.") 24:10-25:21.) As a result of these settlements, only CEA claims remain in this action: those based on trader-based manipulation under direct and vicarious liability theories against Rabobank, and those based on persistent suppression under direct, vicarious liability, and aiding and abetting theories against UBS. (Corrected 4AC ¶¶ 668-705.)

As part of briefing the motion for class certification, the parties have submitted reports and deposition testimony from eight experts. Seven of these experts' opinions have prompted Daubert motions. Rabobank has moved to exclude the opinions of (1) Dr. H. Nejat Seyhun, (2) Dr. Janet Netz, (3) Mr. Craig Beevers, and (4) Mr. Eric Miller. Exchange plaintiffs have moved to exclude the opinions of (5) Dr. Robert Willig, (6) Dr. Christopher Culp, and (7) Dr. R. Glenn Hubbard. We first consider the Daubert motions before proceeding to the motion for class certification itself. 11

1. Rabobank's Daubert Motions

1.1. Dr. Seyhun

Exchange plaintiffs offer two reports from Dr. H. Nejat Seyhun: (1) an initial report dated February 2, 2017 (Decl. of David Kovel ex. G., July 10, 2017, ECF No. 2071); and (2) a rebuttal report dated May 3, 2017 (Decl. of Jefferson Bell ex. 5, June 30, 2017, ECF No. 2008). We refer to these as the Seyhun Initial Report and the Seyhun Rebuttal Report. Across these two reports, Dr. Seyhun's opinions relate primarily to two subjects: (1) the determination of what LIBOR, and each bank's LIBOR submissions, would have been but-for the alleged manipulative conduct, and (2) the determination of what effect changes in LIBOR have on the trading prices of EDF contracts and options. Rabobank does not challenge Dr. Seyhun's qualifications, and we agree that Dr. Seyhun is well qualified to offer these opinions. 12

1.1.1. Models of But-For Published LIBOR and But-For LIBOR Submissions

Dr. Seyhun first opines that certain statistical and mathematical analyses can demonstrate that the defendants' alleged conduct impacted LIBOR and calculate the amount of impact. (Seyhun Initial Report *474 ¶ 12.) To support this conclusion, Dr. Seyhun offers several models for estimating but-for LIBOR and each bank's but-for LIBOR submissions as evidence that these statistical and mathematical analyses are available. ( E.g. , Seyhun Initial Report ¶ 87.) In his initial report, Dr. Seyhun offers two methodologies that involve first the calculation of but-for published LIBOR and then the derivation of each bank's but-for LIBOR submission from but-for published LIBOR. In his rebuttal report, Dr. Seyhun offers two additional models that are entirely distinct from the two models presented in his initial report. In each of these models, Dr. Seyhun first calculates but-for LIBOR submissions for each panel bank and then applies the BBA's trimming methodology-which excludes the top quartile of submissions and the bottom quartile of submissions and averages the remaining submissions-in order to determine but-for published LIBOR.

The two models presented in Dr. Seyhun's initial report each begin with the calculation of but-for published LIBOR. Dr. Seyhun models the mathematical relationship between LIBOR and a reference rate on a daily basis during a "clean period" using an ordinary least squares (OLS) regression, 13 and uses that relationship to predict LIBOR over the Class Period. Though Dr. Seyhun varies the reference rate and clean period at various points in his report, 14 Dr. Seyhun primarily uses the ICAP-Ask rate 15 as the reference rate and the aggregated period from 2000 through 2004 combined with 2013 as the clean period. (Seyhun Initial Report ¶¶ 70-78, tbl.2.2.) Dr. Seyhun defines LIBOR "artificiality" to be the difference between his calculated but-for published LIBOR and actual published LIBOR and tests the statistical significance of the artificiality during the Suppression Period (but not Period 0) using another regression. (Seyhun Initial Report tbl.2.3.)

Dr. Seyhun also estimates but-for LIBOR using the ICAP-Ask rate as the reference rate in a second regression. Otherwise identical, this second ICAP-Ask model uses a clean period of 2000 through 2002 aggregated with 2013. Dr. Seyhun presents an analysis of the statistical significance of his artificiality estimates *475 based on this specification (Seyhun Initial Report tbl.2.4), but does not present the results of the regression itself and does not reference this 2000-2002 plus 2013 regression elsewhere in his initial report.

Dr. Seyhun then derives, for each panel bank, but-for LIBOR submissions from the first ICAP-Ask-based model of but-for LIBOR using two different methods. 16 Under the first method (the "Relative Artificiality" model), Dr. Seyhun assesses the "relative artificiality" of a bank's submission by calculating the difference between a bank's actual submission and actual published LIBOR. If the bank's submission was not trimmed from the calculation of published LIBOR, the relative artificiality equals the difference between published LIBOR and the bank's LIBOR submission. If the bank's submission was trimmed from the calculation of published LIBOR by virtue of being higher than the third quartile submission (or lower than the first quartile submission), Dr. Seyhun limits the bank's relative artificiality to the relative artificiality of the highest (or lowest) submission that was retained in the actual LIBOR calculation. This "relative artificiality" in the bank's submission is then added to overall LIBOR artificiality to estimate the total extent of artificiality in the bank's submission. (Seyhun Initial Report ¶¶ 92-94, figs.1E1-1E17.)

Under the second method (the "CDS Spread" model), Dr. Seyhun first estimates, for each panel bank, the relationship between the bank's LIBOR submission and the spreads on the bank's credit default swaps (CDS). 17 Dr. Seyhun uses a regression specified as the natural logarithm of the LIBOR submission against the natural logarithm of the CDS spread and a set of "dummy" indicator variables corresponding to each day, over a clean period from May 18, 2010 to December 31, 2015. 18 Then, for each bank, Dr. Seyhun calculates the percentage deviation of the bank's CDS spread from the average CDS spread across all panel banks and uses that deviation to calculate the percentage difference between the bank's but-for LIBOR submission from published LIBOR, and applies that difference to but-for LIBOR to calculate the bank's but-for LIBOR submission. (Seyhun Initial Report 95-102.)

In his third model (his rebuttal report's first model), which we refer to as the "Rebuttal Period 0 Model," Dr. Seyhun reviews documents to identify "events" of trader-based manipulation and determines whether the event has "a clear directionality," finding 63 "events" over 45 different days. (Seyhun Rebuttal Report ¶ 488.) Independently, to estimate each bank's but-for *476 LIBOR submission, Dr. Seyhun regresses each bank's LIBOR submission on a given day against the ICAP-Ask rate, a one-day lag of the ICAP-Ask rate, 19 a set of dummy variables for each month over Period 0, a first set of interaction terms between the monthly dummy variables and the ICAP-Ask rate, and a second set of interaction terms between the monthly dummy variables and the lagged ICAP-Ask rate. (Seyhun Rebuttal Report ¶¶ 489-91.) That is, Dr. Seyhun estimates, for each panel bank on a given day, the bank's but-for LIBOR submission based on, among other variables, that day's ICAP-Ask rate and the prior day's ICAP-Ask rate. From these but-for LIBOR submissions, Dr. Seyhun then calculates but-for LIBOR by applying the BBA's trimming methodology. Comparing the results from this model and his documentary analysis, Dr. Seyhun concludes that there is "confirmed manipulation by at least one bank on 45 different occasions." (Seyhun Rebuttal Report ¶ 494.)

In his fourth model (his rebuttal report's second model), which we refer to as the "Rebuttal Suppression model," Dr. Seyhun conducts another regression in order to estimate each panel bank's but-for LIBOR submissions. He regresses, using an aggregated clean period of January 1, 2004 through August 7, 2007 and May 18, 2010 through December 31, 2012, 20 each bank's actual LIBOR submission against (1) the ICAP-Ask rate, (2) a one-day lag of the ICAP-Ask rate, (3) the one-month overnight indexed swap (OIS) rate, (4) the standard deviation of the panel banks' CDS spreads, and (5) the difference between the bank's CDS spread and the panel banks' mean CDS spread. Dr. Seyhun then calculates "artificiality" during the Suppression Period as the difference between a bank's actual submission and the bank's but-for submission estimated using his five-variable regression. As in his initial report, Dr. Seyhun also tests the statistical significance of his measured artificiality.

Rabobank contends that Dr. Seyhun's models, and his opinions based on these models, are unreliable because (1) the four models yield directionally inconsistent results; (2) the four models produce implausible findings of artificiality, including substantial artificiality during purported "clean" periods; (3) the ICAP-Ask models and Rebuttal Period 0 models produce estimates of artificiality inconsistent with Dr. Netz's opinions and Exchange plaintiffs' allegations; and (4) his Relative Artificiality and CDS Spread models of but-for LIBOR submissions are reliant on unreliable estimates of but-for LIBOR improperly estimated using 2013 as part of the clean period. We consider each argument in turn.

1.1.1.1. Internal Inconsistencies

Rabobank, relying on Dr. Hubbard's opinions, contends that Dr. Seyhun's methodologies are internally inconsistent. According to Rabobank and Dr. Hubbard, across Dr. Seyhun's four models of but-for *477 LIBOR (the two ICAP-Ask based models presented in the initial report, the Rebuttal Period 0 model, and the Rebuttal Suppression model), at least one model finds upward manipulation on 65% of days in Period 0 and at least one model finds downward manipulation of LIBOR on 89% of days in Period 0. Similarly, across Dr. Seyhun's four models of but-for LIBOR submissions (the Relative Artificiality and CDS Spread models based on the first ICAP-Ask-based model of but-for LIBOR, the Rebuttal Period 0 model, and the Rebuttal Suppression model), at least one model finds upward manipulation of Rabobank's LIBOR submission on 95% of days in Period 0 and at least one model finds downward manipulation on 84% of days in Period 0.

Because the sum of each set of figures is greater than 100%, Rabobank asserts that this "mathematical impossibility" establishes the unreliability of Dr. Seyhun's models. (Rabobank Seyhun Mem. 8; Hubbard Rebuttal Report ¶ 17 & n.54.) Exchange plaintiffs do not dispute Rabobank's cited statistics, but contend that those statistics are meaningless because Dr. Seyhun's models are presented as alternatives and their results therefore "are not meant to be combined." (Exch. Pls.' Seyhun Opp'n 11.)

Rabobank's "mathematical impossibility" argument is not necessarily persuasive, as the 100% threshold on which the argument relies is of little significance here. An expert's models are not necessarily questionable if that percentage exceeds 100%; the extent to which the models disagree matters more than the mathematical sum of certain results that they produce. For example, if an expert presented two models, one of which found upward manipulation on days amounting to 51 days of a 100-day class period and the other of which found downward manipulation on the remaining 49 days but also one day in which the first model found upward manipulation, that one day of conflicting results does not necessarily call into question the entirety of the models. Conversely, an expert's models may be questionable even if the relevant percentages did not exceed 100%. 21 Further, Dr. Seyhun's models are structured differently-at least in part-in order to reflect differences between Period 0 and the Suppression Period. (Seyhun Initial Report ¶¶ 81-102; Seyhun Rebuttal Report ¶¶ 487-500.)

Nonetheless, in this case, the inconsistencies in the results produced by Dr. Seyhun's models do call into question the models' reliability. Even when intended to measure artificiality during the same period of time, Dr. Seyhun's models produce directionally inconsistent results a substantial portion of the time. Even though all three models are intended to estimate artificiality in Rabobank's LIBOR submissions during Period 0, the Rebuttal Period 0 model produces results directionally inconsistent 22 from those produced by and the CDS Spread and Relative Artificiality models on 49% and 40% of days in Period 0, respectively. (Hubbard Rebuttal Report ¶ 20, tbl.1.) Similarly, the Rebuttal Period 0 model produces directionally inconsistent estimates of artificiality in LIBOR from the two ICAP-Ask-based models of but-for LIBOR on 44% and 42% of the days in *478 Period 0. (Hubbard Rebuttal Report ¶ 24, tbl.2.)

Accordingly, Exchange plaintiffs cannot seriously dispute that Dr. Seyhun's models produce meaningfully different estimates of artificiality, in directionally inconsistent ways. (Hubbard Rebuttal Report exs. 21A-21Q, 22.) Rabobank has not "manipulated" Dr. Seyhun's models as Exchange plaintiffs suggest by comparing the results produced by one model to those produced by other models; rather, Rabobank identified pervasive inconsistencies. Even if the models are formally designated "alternatives," each is nonetheless intended to estimate the same measures: the amount of artificiality in LIBOR and each panel bank's LIBOR submissions. Some minor disagreement between the models would naturally be expected, but Dr. Seyhun's models produce internally inconsistent results on more than half of days. Inconsistencies of this magnitude indicate a lack of reliability. See Lippe , 99 Fed.Appx. at 279 ("[The expert] failed to explain why her [two] analyses did not yield similar results-after admitting that they should."); In re Fed. Home Loan Mortg. Corp. (Freddie Mac) Sec. Litig. , 281 F.R.D. 174 , 181 (S.D.N.Y. 2012) ("[The expert's] analysis changed so many times in important ways and was so internally inconsistent that I found it unreliable and unpersuasive."). Dr. Seyhun has not made a similar concession, but his models are, at bottom, intended to measure the same things-artificiality in LIBOR and LIBOR submissions.

Exchange plaintiffs argue that we should reject Lippe and In re Freddie Mac , contending that we should instead follow U.S. Information Systems, Inc. v. IBEW Local Union No. 3 , 313 F.Supp.2d 213 , 236 (S.D.N.Y. 2004), which held that "purported inconsistencies in [an expert's] methodology go to the weight, not the admissibility of his testimony." (Exch. Pls.' Seyhun Opp'n 10 n.9.). However, these cases establish a distinction between inconsistencies within an expert's application of a single methodology on the one hand and inconsistencies across an expert's multiple methodologies on the other. U.S. Information Systems establishes at most that the former was not a basis for exclusion in that case, see 313 F.Supp.2d at 235-36 (addressing an expert's "inconsistent application of methodology"). 23 By contrast, Lippe , Freddie Mac , and Dr. Seyhun's multitude of models all present the latter issue. See Lippe , 99 Fed.Appx. at 279 (identifying an expert's failure "to explain why her [two alternative] analyses did not yield similar results"); Freddie Mac , 281 F.R.D. at 179-81 (finding an expert's two event studies "internally inconsistent"). In sum, we conclude that the internally inconsistent nature of Dr. Seyhun's models seriously undermine their reliability.

1.1.1.2. Implausible Artificiality Findings

Next, Rabobank argues that Dr. Seyhun's ICAP-Ask-based models of but-for LIBOR and his Rebuttal Period 0 model are unreliable because the three models identify artificiality in published LIBOR on a substantial percentage of days during the models' respective clean periods. Specifically, the first ICAP-Ask-based model identifies artificiality on 57% of days between 2000 and 2004, the second ICAP-Ask-based model identifies artificiality on 39% of days between 2000 and 2002, and the Rebuttal Period 0 model identifies artificiality on every single day in Period 0. (Rabobank Seyhun Mem. 9.)

Rabobank's statistics are not entirely fairly presented, as they represent percentages of only certain portions of the clean periods underlying the three models *479 rather than the entireties of the clean periods. Nonetheless, restating those percentages slightly to represent percentages of the entire clean periods, Dr. Seyhun's models continue to find significant artificiality during clean periods. The first ICAP-Ask-based model identifies artificiality on 50% of clean-period days, 24 the second ICAP-Ask-based model identifies artificiality on 29% of clean-period days, 25 and the Rebuttal Period 0 identifies artificiality on at least 58% of clean-period days. 26

As to the ICAP-Ask-based models of but-for LIBOR, Exchange plaintiffs again do not dispute Rabobank's statistics, but contend that the clean periods in those models "were not 100% clean periods," "merely cleaner periods." (Exch. Pls.' Seyhun Opp'n 12 n.11.) Accordingly, they reassure us, the ICAP-Ask-based models retain validity because they find comparatively more artificiality during the Class Period than during the clean periods.

We seriously question whether a period with artificiality on half of days, or even 29% of days, may be fairly characterized as a "clean" period. As an initial matter, this defense is plainly inconsistent with Exchange plaintiffs' theory of the case, which does not extensively allege manipulation between 2000 and 2004. But even had Exchange plaintiffs done so, the reliability of Dr. Seyhun's regressions-ultimately intended to gauge the relationship between LIBOR and ICAP-Ask absent the defendants' challenged conduct-would be seriously undermined by their incorporation of data pervasively affected by that very conduct. Cf. Fogarazzo v. Lehman Bros., Inc. , 263 F.R.D. 90 , 105 & n.124 (S.D.N.Y. 2009) (expressing skepticism regarding an expert's use of control periods that had been affected by the defendants' challenged conduct). Moreover, this belated rationalization is belied by Dr. Seyhun's opinions. In offering his ICAP-Ask-based models of but-for LIBOR in his initial report, Dr. Seyhun specifically opined that " [b]y definition , the estimation period should be free from manipulation" and suggested that a clean period incorporating the period between May 18, 2010 and December 31, 2012 was inappropriate because "there is some evidence that suggests that some manipulation may have continued into 2012 after the Class Period." (Seyhun Initial Report ¶ 72, n.32 (emphasis added).) 27

Our confidence in the first ICAP-Ask-based model is further undermined by the fact that in addition to finding substantial artificiality between 2000 and 2004, it apparently finds no significant artificiality during Period 0. (Hubbard Initial Report *480 ¶ 125, ex. 10.2.). That is, this model identifies significant LIBOR artificiality when Exchange plaintiffs have not alleged manipulation, and no significant LIBOR artificiality when they have so alleged. While sporadic findings of artificiality in clean periods may be attributable to statistical noise, this exactly-backwards relationship suggests either an unreliable model or a deeper flaw in Exchange plaintiffs' theory of the case. Because the Relative Artificiality and CDS Spread models of but-for LIBOR submissions are reliant on the first ICAP-Ask-based model's estimates of but-for LIBOR, the reliability of those models is necessarily called into question as well.

As to the Rebuttal Period 0 model, Exchange plaintiffs respond first that Dr. Seyhun's Rebuttal Period 0 model "predicts artificiality [only] on days where there exists documented evidence that is consistent in direction with the model's prediction" (Exch. Pls.' Seyhun Opp'n 11), and second that the Rebuttal Period 0 Model "breaks up Period 0 into separate control and treatment periods" based on "documented evidence of TBM," (Exch. Pls.' Seyhun Opp'n 16).

Exchange plaintiffs' first response is little more than definitional sleight-of-hand. Rabobank's contention is that the Rebuttal Period 0 model finds mathematical artificiality on 100% of days in Period 0-that is, the Rebuttal Period 0 model calculates a but-for LIBOR that is different from actual published LIBOR. By contrast, Exchange plaintiffs' response defines "artificiality" to mean mathematical artificiality supported by "corroborating evidence" based on Dr. Seyhun's assessment of documents.

This terminological slipperiness is unavailing. In contending that Dr. Seyhun's models are in fact capable of identifying trader-based manipulation, Exchange plaintiffs argue that "on days that documents do not show evidence of manipulation, the TBM Model finds positive artificiality only 51% of the time" and "negative artificiality only 49% of the time." (Exch. Pls.' Seyhun Opp'n 17 n.18 (emphasis added) (citing Seyhun Rebuttal Report ¶ 493).) But this statement is wholly irreconcilable with their contention now that the Rebuttal Period 0 "predicts artificiality [only] on days where there exists documented evidence that is consistent in direction with the model's prediction." (Exch. Pls.' Seyhun Opp'n 11.) Accepting this latter statement as true, then on days where no documents show evidence of manipulation, the TBM Model should find positive artificiality 0% of the time and negative artificiality 0% of the time. But it does not.

Ultimately, Dr. Seyhun's Rebuttal Period 0 Model finds mathematical artificiality on 100% of days in Period 0, and we conclude that this implausible finding of (mathematical) artificiality undermines the reliability of the Rebuttal Period 0 Model. Exchange plaintiffs' attempt to obfuscate this reality by shifting the definition of "artificiality" to include a requirement of documentary corroboration does not affect our conclusion regarding the unreliability of Dr. Seyhun's methodology. Indeed, accepting this revised definition of "artificiality," Dr. Seyhun's models would be wholly incapable of identifying instances of TBM-caused artificiality. The documentary corroboration requirement becomes the only filter separating true "artificiality" from the mathematical artificiality identified by the Rebuttal Period 0 Model on 100% of days in Period 0, even though the days may otherwise be difficult to distinguish. 28 Exchange plaintiffs' dual contentions-that *481 the Rebuttal Period 0 model is capable of identifying trader-based manipulation independent of documentary evidence on the one hand (based on its findings of mathematical artificiality) but that the Rebuttal Period 0 model does not identify artificiality on 100% of days in Period 0 (because it requires documentary evidence)-cannot be reconciled.

Nor do we find much comfort in Exchange plaintiffs' assurances that Dr. Seyhun's Rebuttal Period 0 model finds positive (mathematical) artificiality on 71% of days with documentary evidence of upward manipulation and finds negative (mathematical) artificiality on 75% of days with documentary evidence of downward manipulation. (Exch. Pls.' Seyhun Opp'n 17 n.18.) Rephrasing slightly, this contention acknowledges that even on days for which Dr. Seyhun has already identified documentary evidence of manipulation, the Rebuttal Period 0 model has an error rate between 25% and 29%.

Exchange plaintiffs' second response, that the Rebuttal Period 0 model in fact divides Period 0 into a control period and a treatment period, is difficult to evaluate based on Dr. Seyhun's report, and Rabobank disputes whether Dr. Seyhun based his regressions only on days without evidence of trader-based manipulation. If Dr. Seyhun did in fact separate days in Period 0 with documentary evidence of trader-based manipulation from days without such evidence and used only the latter as a control period as plaintiffs assert, his report is conspicuously devoid of any indication to that effect. (Seyhun Rebuttal Report 487-94.) Nonetheless, we accept counsel's representation at oral argument that Dr. Seyhun did so separate (Hr'g Tr. 5:2-6:8), and conclude that this second defense is also unpersuasive. Regressions relying on piecemeal clean periods have been described as "unorthodox," and at least one court has noted that the selection of a clean period wholly independent of periods involving challenged conduct is recommended practice. See Fogarazzo , 263 F.R.D. at 105 . Indeed, this choice of clean period becomes especially difficult to justify given that Exchange plaintiffs assert that there likely exist additional instances of trader-based manipulation in Period 0 that could be identified through additional discovery, which would taint days in the clean period-a point acknowledged by counsel at oral argument. (Hr'g Tr. 6:3-9; Exch. Pls.' Seyhun Opp'n 13.)

1.1.1.3. Inconsistencies with Dr. Netz

Rabobank casts further doubt on the reliability of Dr. Seyhun's methodologies by identifying substantial inconsistency between Dr. Seyhun's findings and Dr. Netz's findings on the existence and direction of manipulation. Rabobank notes that, when compared to Dr. Netz's findings of manipulation of Rabobank's LIBOR submissions in a particular direction, Dr. Seyhun identifies LIBOR artificiality in the opposite direction 52% of the time using his first ICAP-Ask-based model of but-for LIBOR and 56% of the time using his Rebuttal Period 0 Model. (Hubbard Initial Report ¶¶ 84-86, ex. 1.1; Hubbard Rebuttal Report ¶¶ 46-47.)

Exchange plaintiffs again do not deny the existence of these inconsistencies, but relying on In re Vitamin C Antitrust Litigation , No. 06 MD 1738 (BMC)(JO), 2012 WL 6675117 , at *4 (E.D.N.Y. Dec. 21, 2012), suggest that any disagreement between Dr. Seyhun and Dr. Netz is merely a "battle of the experts" that need not be resolved now. However, we are unpersuaded that fundamental disagreements between a party's own experts can be considered a mere "battle of the experts." See, e.g. , Deutsch v. Novartis Pharm. Corp. , 768 F.Supp.2d 420 , 469 (E.D.N.Y. 2011)

*482 (identifying inconsistencies between two of the same party's experts as a basis for exclusion).

Further, Dr. Netz's opinions regarding specific instances of trader-based manipulation largely overlap with the allegations of trader-based manipulation set forth in the Exchange plaintiffs' operative complaint. ( Compare Netz Initial Report ex. 1, with, e.g. , Corrected 4AC ¶¶ 157-79 (Barclays allegations); id. ¶¶ 194-97 (Rabobank allegations); id. ¶ 227 (Deutsche Bank allegations).) Consequently, to the extent Dr. Seyhun's opinions contradict Dr. Netz's opinions regarding specific instances of trader-based manipulation and the direction of that manipulation, they contradict Exchange plaintiffs' allegations as well. Such a misfit also indicates a lack of reliability.

Exchange plaintiffs further assure us that any inconsistencies between Dr. Seyhun's opinions and Dr. Netz's opinions (and, by extension, the allegations in the complaint) are of no concern because "where documentary evidence is limited, an initial review of that evidence may yield a result that is directionally inconsistent with the results of a statistical analysis." (Exch. Pls.' Seyhun Opp'n 12-13.) For instance, they suggest, if on a given day, one bank manipulated its submission downward and two banks manipulated their submissions upward, but only documentary evidence of the first bank's manipulation were available, Dr. Seyhun's models could correctly show upward manipulation in LIBOR even though such a result would be seemingly inconsistent with the available documentary evidence (of only the first bank's downward manipulation). (Exch. Pls.' Seyhun Opp'n 13 n.13.)

The possibility of undiscovered instances of manipulation explains away little of the inconsistency between Dr. Seyhun and Dr. Netz, as further documentary evidence of manipulation by additional banks cannot explain away directional inconsistency between documentary evidence of trader-based manipulation by one bank as identified by Dr. Netz (and as alleged by Exchange plaintiffs) and artificiality in that bank's LIBOR submissions as calculated by Dr. Seyhun's models. Accordingly, to the extent artificiality in LIBOR as calculated by Dr. Seyhun's models shares the same direction as artificiality in LIBOR submissions, inconsistency between Dr. Netz's findings (and Exchange plaintiffs' allegations) and the former implies inconsistency between Dr. Netz's findings and the latter as well.

Under Dr. Seyhun's Relative Artificiality model, the direction of artificiality in each bank's LIBOR submissions will generally be the same as the direction of artificiality in overall published LIBOR. Indeed, when a bank's LIBOR submission is within the interquartile range and is therefore retained in calculating LIBOR, Dr. Seyhun's Relative Artificiality model will find artificiality in the bank's LIBOR submission identical in both direction and magnitude to artificiality in LIBOR. Similarly, under the CDS Spread Model of Rabobank's but-for LIBOR submissions, Dr. Seyhun generally finds that Rabobank manipulated its submissions upward throughout Period 0. Figure 1F13 in the Seyhun Initial Report, displaying the CDS Spread Model's results for Rabobank, generally shows Rabobank's actual LIBOR submissions to be above Rabobank's but-for LIBOR submissions during Period 0. By contrast, of the 32 instances of TBM identified by Dr. Netz associated with Rabobank, she finds that only 17 (or 53%) involved upward manipulation. 29 (Netz Initial *483 Report ex. 1; Netz Rebuttal Report ex. 26.) That is, Dr. Netz reaches an inconsistent conclusion on 47% of days. Further, Dr. Seyhun acknowledges inconsistency between the direction of artificiality in a bank's LIBOR submissions suggested by documentary evidence-also the basis for Dr. Netz's opinions-and the direction of artificiality calculated by the Rebuttal Period 0 Model: two of the exhibits in the Seyhun Rebuttal Report identify only "Documented Trader Manipulation Events Consistent with [the Rebuttal] Period 0 Model," suggesting the exclusion of inconsistent events. (Seyhun Rebuttal Report ¶ 493, exs. 1-2.)

In sum, none of these inconsistencies between Dr. Seyhun's various models of but-for LIBOR submissions and Dr. Netz's document-based identification of manipulation can be explained away by the specter of additional documentary evidence of manipulation by additional panel banks. Accordingly, these inconsistencies further erode our confidence in the reliability of Dr. Seyhun's models.

1.1.1.4. Inclusion of 2013 in Clean Periods

Rabobank argues that Dr. Seyhun's inclusion of 2013 in his clean periods is improper cherry-picking, as the ICAP-Ask-based models of but-for LIBOR show no statistically significant artificiality during Period 0 when 2013 is removed from the clean period. The inclusion of 2013 is problematic, Rabobank argues, because the relationship between published LIBOR and ICAP-Ask differs between the two discontinuous pieces of Dr. Seyhun's clean periods-the spread averages 2.7 basis points between 2000 and 2004, but averages negative 1.4 basis points in 2013 (Rabobank Seyhun Mem. 13; Hubbard Initial Report ¶ 115, ex. 7.1)-and because the ICAP-Ask rate in 2013 exhibits data-quality problems in that it remains entirely unchanged for more than eight consecutive months. (Rabobank Seyhun Reply 3 n.8; Hubbard Initial Report ¶ 116, ex. 8.1.)

Exchange plaintiffs admit that the spread between LIBOR and the ICAP-Ask rate in 2000-2004 (and 2000-2002) differs from the spread in 2013, do not dispute the ICAP-Ask data-quality issue, and do not appear to dispute Dr. Hubbard's findings that the Initial ICAP-Ask models identify no statistically significant artificiality when 2013 is removed from the respective clean periods. (Exch. Pls.' Seyhun Opp'n 14-15). Rather, they argue that the difference in spread is "inconsequential" because the spread is "absorbed" into the intercept term of Dr. Seyhun's regressions. (Exch. Pls.' Seyhun Opp'n 15; Seyhun Rebuttal Report ¶ 463.)

This contention is unavailing. In assessing the relationship of LIBOR and the 3-month ICAP-Ask rate, Dr. Seyhun conducts an Ordinary Least Squares (OLS) regression 30 specified as follows:

LIBOR = a + ß. - (ICAP-Ask) + e

(Seyhun Report tbl.2.2.) In an OLS regression of this form, a represents the intercept, ß represents the coefficient capturing the relationship between ICAP-Ask and LIBOR, and e represents the error term. In turn, ß is calculated as (1) the covariance of dependent variable (LIBOR) and the single explanatory variable (ICAP-Ask), divided by (2) the variance of the explanatory variable (ICAP-Ask). The intercept term, a, is then calculated as the mean of the dependent variable (LIBOR) minus the product of the coefficient ß and the mean of explanatory variable (ICAP-Ask). The different absolute relationship between LIBOR and ICAP-Ask in 2000-2004 and in 2013 will affect the covariance of the two series, which in turn impacts the calculated beta. Exchange plaintiffs *484 are correct in that a is also impacted by this difference in spread, since a is, in part, a function of ß. Nonetheless, unless ß equals exactly 1, the spread between LIBOR and ICAP-Ask will never be differenced out in its entirety. That is, not only is the changing spread not entirely accounted for by the intercept term, the reason it is not so "absorbed" is because the changing spread impacts ß, the estimate of how much but-for LIBOR should change given a change in the ICAP-Ask rate.

That is, we know that the inclusion of 2013 in Dr. Seyhun's clean periods is a significant driver of his results, even though the inclusion of 2013 in the clean periods not only introduces a significant data-quality issue but also introduces a qualitatively different relationship between ICAP-Ask and published LIBOR, a difference that is only partially absorbed into the intercept term. Absent an affirmative justification for adding 2013 to the clean periods and a more robust explanation for why such an addition does not skew the results, we conclude that Dr. Seyhun's inclusion of 2013 strongly suggests cherry-picking and renders the ICAP-Ask-based models of but-for LIBOR (and the Relative Artificiality and CDS Spread models of but-for LIBOR submissions) unreliable. 31

1.1.2. Effect of Changes in LIBOR on EDF Prices

Dr. Seyhun opines that "statistical regression or other mathematical analyses" are available to show that "artificial LIBOR caused Eurodollar futures (and options) contract prices to be artificial," and that further analyses are available to "estimate the approximate amounts of such artificiality." (Seyhun Initial Report ¶ 12.b (emphasis added).) Dr. Seyhun bases this opinion on numerous regression models, from which he derives a number of "impact factors" that purport to measure, on a daily basis, how a change in LIBOR would have been reflected in EDF prices.

Across his two reports, Dr. Seyhun calculates 31 different values representing the "impact" of a change in 3-month LIBOR on each of the 44 Eurodollar contracts traded on the Chicago Mercantile Exchange: 10 in his initial report and 21 in his rebuttal report. (Seyhun Initial Report tbls.5, 10.2; Seyhun Rebuttal Report tbls.3, 6, 9, 12.) Dr. Seyhun also calculates corresponding sets of impact factors for changes in 1-month LIBOR and 6-month LIBOR. ( E.g. , Seyhun Rebuttal Report tbls.2, 4.) However, because Dr. Seyhun focuses his opinions on measuring the effect of changes in 3-month LIBOR, we so focus as well.

Broadly speaking, these 31 models vary in three primary ways: (1) whether the 3-month LIBOR series is contemporaneous or shifted backwards by one day ( i.e. , for each day, using the change in LIBOR between the day in question and the day after , rather than the change in LIBOR from the day before to the day in question); (2) whether changes in the implied 3-month forward rate are included as explanatory variables; (3) whether, and how many, lags of explanatory variables are used. 32

*485 Initial Report Table 5 presents the results of three models, each using shifted LIBOR and omitting as an explanatory variable changes in the implied 3-month forward rate. 33 The first model regresses changes in EDF prices on changes in LIBOR, with no lags. The second model regresses changes in EDF prices on lagged changes in EDF prices and changes and lagged changes in LIBOR, using the number of lags suggested by the Bayesian information criterion. The third model is like the second model, but uses the number of lags suggested by the Akaike information criterion. 34 Initial Report Table 10.2 presents the results of seven models, each using next-day LIBOR and including as an explanatory variable changes in the implied 3-month forward rate. Each model regresses changes in EDF prices on lagged changes in EDF prices, changes and lagged changes in 3-month LIBOR, and changes and lagged changes in the implied 3-month forward rate. The number of lags varies from 0 to 6, with one model for each number.

Dr. Seyhun's rebuttal report introduces 21 models intended to supersede, at least in part, the 10 models presented in his initial report. Rebuttal Report Table 3 presents the results of four models, three of which use the same regression specification as the three models presented in Initial Report Table 5. However, the results here differ from those presented in the Initial Report Table 5 because Dr. Seyhun has corrected the data processing error attributable to "rollover" days made in his initial report. By contrast, the fourth model regresses changes in EDF prices on changes in today's LIBOR and changes in next-day LIBOR. Dr. Seyhun uses the impact factors produced by one of these models for his damages estimates, though he does not specify which one. (Seyhun Rebuttal Report ¶ 86.) Rebuttal Report Table 6 presents the results of seven models, specified exactly as the seven models reported in Initial Report Table 10.2 (including the use of next-day LIBOR). As with the relationship between Rebuttal Report Table 3 and Initial Report Table 5, the results here differ from those presented in the corresponding Initial Report Table 10.2 because Dr. Seyhun has corrected the data processing error attributable to "rollover" days. Rebuttal Report Table 9 presents the results of three models, specified identically to Initial Report Table 5, but using contemporaneous LIBOR rather than next-day LIBOR and correcting the data-processing error. Similarly, Rebuttal Report Table 12 presents the results of seven models, specified identically to Initial Report Table 10.2 (and Rebuttal Report Table 6), but also using contemporaneous LIBOR and correcting the data-processing error.

Rabobank offers several criticisms of these models: (1) that they cannot be properly interpreted to support causation; (2) that they fail to account for intraday movements in EDF prices; (3) that they fail to account minimum price increments in *486 EDF prices; and (4) that they are contradicted by Dr. Culp's findings that EDF prices exhibit little reaction when LIBOR is published. 35

1.1.2.1. Correlation, Not Causation

Rabobank and Exchange plaintiffs primarily dispute whether Dr. Seyhun's models can be properly interpreted to suggest that changes in LIBOR cause determinable changes in EDF prices 36 -Dr. Seyhun's ultimate opinion for which he relies on these models.

As to the 11 models relying on shifted (next-day) LIBOR, which, according to Dr. Seyhun, establishes that changes in tomorrow 's LIBOR cause changes in EDF prices today , we reject them as unreliable at the threshold. The causal relationship purportedly supported by these regressions defies common sense: needless to say, a phenomenon must precede any effects that it causes, and Exchange plaintiffs offer no remotely cogent theory as to how changes in LIBOR manage to travel backward in time in order to affect EDF prices in the past. 37 Dr. Seyhun acknowledges this feature of his models, but contends that his methodology is proper because next-day LIBOR better incorporates "any economic event that happens during most of the U.S. business hours" and "U.S. business developments." (Seyhun Rebuttal Report ¶ 84; Exch. Pls.' Seyhun Opp'n ¶ 20.) That is, Dr. Seyhun asserts, using next-day LIBOR "captures the fact that Eurodollar futures prices can at times react earlier to interest-rate sensitive information before LIBOR actually changes." (Seyhun Rebuttal Report ¶ 87.) In offering these defenses of Dr. Seyhun's models, Exchange plaintiffs have apparently lost sight of the ultimate proposition that Dr. Seyhun set out to prove: that changes in LIBOR cause determinable changes in EDF prices. Exchange plaintiffs' references to "economic events," "business developments," and "interest-rate sensitive information" admit only one fair interpretation: namely, as a concession that those events, those developments, and that information confound the relationship between LIBOR and EDF prices presented in Dr. Seyhun's models, which offer no means of distinguishing changes in EDF prices caused by changes in LIBOR from other conditions causing changes in both EDF prices and LIBOR.

These confounding variables loom large, too, over Dr. Seyhun's remaining models relying on contemporaneous LIBOR. The *487 fact that these models do not rely on the assumption that changes in LIBOR have retrospective effect does not remedy the more glaring error that Rabobank has identified and that Exchange plaintiffs have conceded. Because Dr. Seyhun's models offer no means of controlling for the effects of economic events and business developments, we conclude that they cannot reliably support Dr. Seyhun's causation opinion. See Bickerstaff v. Vassar Coll. , 196 F.3d 435 , 449-50 (2d Cir. 1999) ; Reed Constr. Data Inc. v. McGraw-Hill Cos., 49 F.Supp.3d 385 , 400 (S.D.N.Y. 2014) ; Bonton v. City of New York , No. 03 Civ. 2833 (SAS), 2004 WL 2453603 , at *3 n.32 (S.D.N.Y. Nov. 3, 2004).

In sum, "[t]he argument post hoc, ergo propter hoc , is not always a strong one; but the argument ante hoc, ergo propter hoc , must be a good deal weake[r]." 7 May 1861, 162 Parl Deb HC (3d ser.) 1677, 1679. Dr. Seyhun's contemporaneous LIBOR models cannot support his opinions regarding causation, as Exchange plaintiffs concede (and, curiously, emphasize) that LIBOR and EDF prices are each driven by business developments and economic events (Exch. Pls.' Seyhun Opp'n 20; Seyhun Rebuttal Report ¶ 84), and Dr. Seyhun's models offer no means of separating changes in EDF prices caused by those events rather than by changes in LIBOR. Dr. Seyhun's next-day LIBOR models fare even worse, not only suffering from the same confounding variable problem but also relying on the unfounded assumption that a change in LIBOR one day can effect changes in EDF prices in the past.

1.1.2.2. Intraday Variations in Price

Rabobank next argues that Dr. Seyhun's models fail to analyze intraday EDF data, opting instead to rely on the daily settlement prices calculated each day at 2:00 p.m. Chicago time. This method is unreliable, Rabobank contends, because it relies upon the unfounded assumption that any artificiality in LIBOR has an impact of constant magnitude over the 23 hours per day during which EDFs are traded. (Rabobank Seyhun Mem. 21-22.) Exchange plaintiffs respond that no analysis of intraday EDF data is necessary "because there is a single LIBOR fix per day and a single LIBOR artificiality per day," and consideration of intraday measurements would be "futile for purposes of demonstrating the impact of LIBOR on EDF prices." (Exch. Pls.' Seyhun Opp'n 22.)

Notably, Exchange plaintiffs do not dispute that Rabobank's assertion that Dr. Seyhun's decision to conduct his analysis at a daily level incorporates the assumption that LIBOR artificiality has a single, constant amount of impact on EDF prices over a given trading day. But this assumption is inconsistent with Dr. Seyhun's assertions that EDFs "trade in efficient markets and react to all relevant information in a timely and complete manner," (Seyhun Rebuttal Report ¶ 124), and Exchange plaintiffs' assertion that "EDF prices continually incorporate all fundamental information" (Exch. Pls.' Seyhun Opp'n 22). And indeed, intraday variations in EDF pricing are of particular relevance to this action. Unlike, for example, the OTC or Lender plaintiffs, whose LIBOR-based instruments are pegged to LIBOR and reset on a periodic basis using published LIBOR on a given day, EDF traders' exposure to changes in LIBOR, even under Exchange plaintiffs' theory, was less periodic and less consistent. The record establishes that named plaintiffs actively opened and closed EDF positions within a single day (Ordover Initial Report sec. VIII), and Exchange plaintiffs themselves have acknowledged that these "in-and-out" traders make up a large proportion of the putative class (Nov. 1, 2017 Hr'g Tr., 12:24-13:11, ECF No. 2341).

*488 Given the importance of intraday trading in assessing the impact of LIBOR artificiality on EDF traders, Dr. Seyhun's failure to account for intraday variation in EDF prices further undermines the reliability of his ultimate opinion that statistical methods are available to determine that impact.

1.1.2.3. Minimum Price Increments ("Tick" Size)

Rabobank further criticizes Dr. Seyhun's model for failing to incorporate "ticks," or the minimum price increments in which EDF contracts trade. Under Rabobank's theory, the impact of trader-based manipulation on published LIBOR was often less than one tick in magnitude, and such a small extent of impact will not move EDF prices by one "tick" even assuming some causal relationship flowing from LIBOR to EDF prices. Therefore, Rabobank argues, Dr. Seyhun's model is further flawed for this failure to incorporate minimum tick sizes. (Rabobank Seyhun Mem. 22-23.) Exchange plaintiffs respond that Dr. Seyhun has in fact offered a formula to account for minimum price increments, and that Dr. Seyhun's failure to incorporate that formula into his models is of "no consequence" because "it involves a logical rather than an empirical argument." (Exch. Pls.' Seyhun Opp'n 22-23.)

Exchange plaintiffs are correct to the limited extent that Dr. Seyhun has indeed offered a formula regarding tick sizes. This formula is, however, wholly unhelpful: Dr. Seyhun opines that the probability that LIBOR artificiality affected EDF prices is equal to the amount of LIBOR artificiality divided by the tick size for the EDF contract in question. (Seyhun Rebuttal Report ¶ 216.) But knowing that EDF prices will be affected, on average, a certain percentage of the time is insufficient; the operative question for class certification purposes is determining which prices were affected and which prices were not. That is, we are ultimately interested in the empirical question of how much EDF prices were affected by artificiality in LIBOR; knowing that EDF prices may or may not, as a matter of logic, have been impacted a certain percentage of the time is not responsive.

Exchange plaintiffs attempt to minimize the importance of tick sizes, reasoning that "[t]he minimum tick size will either nullify the effect of LIBOR artificiality on that day, or amplify it, based on the tipping point of the minimum tick size." (Exch. Pls.' Seyhun Opp'n 23.) We agree with this analysis of the impact of minimum tick sizes; indeed, it is the root of our concern regarding Dr. Seyhun's failure to consider tick sizes: the difference between amplification or nullification of LIBOR artificiality, i.e. , some impact or no impact whatsoever , is difficult to overstate. Knowing which of the two will occur is of critical importance in assessing the impact of LIBOR manipulation on EDF prices, and Dr. Seyhun's models offer no aid in making that assessment. While we could conceive of an additional rounding step that could be incorporated into Dr. Seyhun's models, it is Exchange plaintiffs' responsibility, and not ours, to provide models that are sufficiently complete so as to be reliable and helpful to the factfinder. Dr. Seyhun's failure to do so further supports the exclusion of his opinions.

1.1.3. Conclusion

Rabobank's motion to exclude Dr. Seyhun's opinions is granted. Dr. Seyhun's ultimate opinions, that classwide methods are available to determine the existence and extent of LIBOR manipulation and to determine the impact of that manipulation on EDF prices, are unreliable. The numerous models that Dr. Seyhun proposes are inadequate to support this conclusion, as Exchange plaintiffs have not established that those models are themselves reliable.

*489 Because Dr. Seyhun's opinions regarding damages rely on his opinions regarding the determination of LIBOR manipulation and resulting impact on EDF prices, we conclude that those opinions should be excluded as well.

Dr. Seyhun's multitude of models of artificiality in LIBOR and LIBOR submissions are rife with inconsistencies: they are internally inconsistent, they are inconsistent with Dr. Netz's opinions, and they are inconsistent with Exchange plaintiffs' own allegations of trader-based manipulation. Dr. Seyhun's models of changes in EDF prices caused by changes in LIBOR, though greater in number, are no more reliable, as they are not only incapable of separating the effect of macroeconomic events but also incapable of actually determining when EDF prices would have changed in light of minimum price increments. We have previously questioned whether plaintiffs would "be able to show that LIBOR suppression of a particular amount would have caused a corresponding, determinable change in [EDF] trading prices," LIBOR VI , 2016 WL 7378980 , at *22, slip op. at *59, and nothing in Dr. Seyhun's reports causes us to reconsider that skepticism.

1.2. Dr. Netz

Exchange plaintiffs present two reports from Dr. Janet Netz: (1) an initial report dated February 2, 2017 (Decl. of Thomas Elrod ex. 1, July 21, 2017, ECF No. 2121); and (2) a rebuttal report dated May 3, 2017 (Decl. of Thomas Elrod ex. 2, July 21, 2017, ECF No. 2121). We refer to these as the Netz Initial Report and the Netz Rebuttal Report. In her reports, Dr. Netz offers: (1) a number of opinions derived from her review of communications between traders; (2) the opinion that manipulation in LIBOR submissions impacted published LIBOR; (3) the opinion that changes in LIBOR cause calculable changes in prices of EDF contracts and options; and (4) a number of opinions regarding the calculation of damages. Based on Dr. Netz's credentials, 38 we conclude that she is generally qualified to offer opinions of this type. We accordingly reject Rabobank's challenges to Dr. Netz's expert qualifications, but address them specifically in assessing Dr. Netz's specific opinions.

1.2.1. Opinions Based on Trader Communications

Dr. Netz devotes a substantial portion of her reports to analyzing documentary evidence of communications between traders, LIBOR submitters, and other individuals affiliated with panel banks. Relying on her interpretation of these documents, Dr. Netz opines that (1) panel banks manipulated their LIBOR submissions; (2) these manipulated LIBOR submissions impacted published LIBOR; (3) panel banks believed that LIBOR manipulation would affect EDF prices. In reaching these opinions, Dr. Netz also concludes that review of documentary evidence "is a common, class-wide method using common evidence that can prove liability." (Netz Rebuttal Report 10.)

At the outset, we note that Exchange plaintiffs and Rabobank dispute the scope of Dr. Netz's opinions. Rabobank asserts that Dr. Netz concludes that certain challenged conduct in fact occurred, and Exchange *490 plaintiffs respond that Dr. Netz "does not opine that Defendants engaged in trader-based conduct or colluded in doing so," but rather opines only that "documents consistent with those allegations are available, and are common to the class." (Exch. Pls.' Netz Opp'n 8 (emphasis omitted).)

Dr. Netz's opinions simply are not limited in the way Exchange plaintiffs suggest. The Netz Initial Report contains numerous conclusions that, based on her review of trader communications, certain challenged conduct in fact occurred and Exchange plaintiffs were in fact impacted. For example, Dr. Netz opines that "[t]he discovery record shows that Defendants manipulated their LIBOR submissions in ways that were inconsistent with the reporting rules established by the BBA," that "Defendants engaged in coordinated trader-based manipulation," and that "Defendants engaged in suppression manipulation." (Netz Initial Report 13, 18, 20.) While Dr. Netz later opines on the availability and applicability of common evidence, for example that "[l]iability can be proven using common, class-wide methods based on common evidence," (Netz Rebuttal Report 4), these subsequent opinions do not supersede her earlier opinions. We consider different aspects of Dr. Netz's opinions in turn.

To the extent Dr. Netz offers her interpretation of trader communications to conclude that manipulation in fact occurred, these opinions are not based on Dr. Netz's "scientific, technical, or other specialized knowledge" as required by Rule 702(a). While Exchange plaintiffs correctly suggest that an expert's testimony may be permissible, for example, "to translate esoteric terminology," United States v. Mejia , 545 F.3d 179 , 190 (2d Cir. 2008), the trader communications that Dr. Netz interprets are clear on their face and "address issues of fact that [the trier of fact] is capable of understanding without the aid of expert testimony," In re Longtop Fin. Techs. Ltd. Sec. Litig. , 32 F.Supp.3d 453 , 460 (S.D.N.Y. 2014). 39 Exchange plaintiffs fail to explain what specialized knowledge Dr. Netz drew upon to conduct her review of trader communications, and "[a]cting simply as a narrator of the facts does not convey opinions based on an expert's knowledge and expertise." Tourre , 950 F.Supp.2d at 675 . Our concern is heightened here, as Dr. Netz offers no explanation of how the communications that she reviewed were selected.

Similarly, to the extent Dr. Netz interprets trader communications to opine on the traders' actual motives and states of mind, these opinions are impermissible. See, e.g. , Highland Capital Mgmt., L.P. v. Schneider , 379 F.Supp.2d 461 , 469-70 (S.D.N.Y. 2005) ; In re Rezulin , 309 F.Supp.2d at 547 . While she may opine on traders and LIBOR submitters' economic incentives as a general matter, those opinions may not extend to the traders and submitters' specific states of mind.

Finally, to the extent that Dr. Netz opines on the existence of documents, these opinions are also excluded. The general proposition that some documents may pertain to all class members is obvious and not insightful, and Exchange plaintiffs offer no explanation why someone with Dr. Netz's expertise and qualifications is needed to offer it. At class certification, the operative question is whether LIBOR manipulation, once established through documents or other evidence, will pertain to all class members; the issue at this stage is not whether those documents exist. Indeed, Exchange plaintiffs offer no authority *491 supporting their position that the "existence [of documents and data regarding defendants' conduct] is an appropriate topic for the opinion of Plaintiffs' expert on class certification," (Exch. Pls.' Netz Opp'n 7), and we find none.

Dr. Netz's opinions that documents exist establishing certain elements of liability, when properly understood, are opinions on the merits that are irrelevant at the class certification stage. An economic expert at class certification may permissibly opine that event x -as established through documents or otherwise-has an economic effect that is common to all class members. But she may not opine that event x in fact occurred, which remains for the trier of fact to determine. Even when such an opinion is formulated as "evidence exists establishing that event x occurred," it is no more relevant at class certification and is no more admissible than a direct opinion on the merits.

Further, even if these opinions were offered at the merits stage, the latter formulation would be no less invasive of the province of the trier of fact. And as a corollary, it would be no more admissible at that stage, either. See Tourre , 950 F.Supp.2d at 675 ("Expert testimony may not usurp the province of the judge to instruct on the law, or of the jury to make factual determinations."). 40 Therefore, even accepting Exchange plaintiffs' characterization that Dr. Netz "opine[s] on whether documents exist that could prove [their] allegation[s] on a common basis," (Exch. Pls.' Netz Opp'n 12), her opinions to that effect remain inadmissible.

In sum, we conclude that Dr. Netz's opinions interpreting documents and commenting on their existence are not admissible. Of course, this holding regarding Dr. Netz's opinions derived from her review of trader communications does not, by itself, preclude her from opining that manipulation of LIBOR submissions impacted LIBOR in a way common to all class members, or from opining that changes in LIBOR impacted EDF prices in a way common to all class members. We next consider whether those opinions are inadmissible for other reasons.

1.2.2. Effect of Manipulation on LIBOR

Dr. Netz subsequently opines that "manipulated LIBOR submissions affected published LIBOR." (Netz Initial Report 29.) As to trader-based manipulation, Dr. Netz first identifies, by reviewing documentary evidence, 163 instances of alleged trader-based manipulation. Dr. Netz determines that in 111 of those instances, the manipulated submission could have affected published LIBOR once the BBA's trimming methodology is considered. In turn, for each of those 111 instances, Dr. Netz assumes that, in the absence of manipulation, the panel bank in question would have made as its LIBOR submission the median submission reported by all panel banks on that day, and concludes that the replacement of a manipulated submission with the median submission would have affected published LIBOR on 65 of those days. Dr. Netz then highlights three specific examples (from the 65 days identified through her documentary analysis), examines the range of LIBOR submissions made on those dates, and explains how the alleged manipulation could have impacted published LIBOR. (Netz Initial Report 29-31.)

Rabobank contends that these opinions are inadmissible because they are not reliable *492 and not helpful to the trier of fact. Specifically, Rabobank asserts that Dr. Netz's analysis is unreliable because it is based on the "arbitrary and speculative assumption" that a panel bank would have submitted the median of all banks' submissions absent trader-based manipulation. (Rabobank Netz Mem. 12.) In opposition, Exchange plaintiffs respond that Dr. Netz's analysis is probative of the "hotly contested" question of whether trader-based manipulation of one bank's submission could have affected published LIBOR, given the trimming of submissions outside the interquartile range. (Exch. Pls.' Netz Opp'n 13-14.) Rabobank implicitly concedes that whether a single bank's LIBOR submissions could have impacted published LIBOR is a germane question, but disputes Exchange plaintiffs' characterization of the issue as "hotly contested" based on the consistency between Dr. Netz's and Dr. Hubbard's opinions on the issue. (Rabobank Netz Reply 4-5.)

We cannot conclude, as Rabobank urges, that Dr. Netz's opinions in this realm are entirely unhelpful. The ability to impact the market, the existence of artificial prices, and the defendant's causation of those prices are elements of a CEA claim, see, e.g. , LIBOR I , 935 F.Supp.2d at 713 ; In re Platinum & Palladium Commodities Litig. , 828 F.Supp.2d 588 , 598 (S.D.N.Y. 2011), and the question of whether its submissions could impact published LIBOR is therefore "a fact in issue," Fed. R. Evid. 702(a).

We find little merit in Rabobank's contention that Dr. Netz's opinion here is irrelevant because she and Dr. Hubbard reach the same conclusion: that Rabobank could have impacted 3-month LIBOR on 46% of days in Period 0. The parties' agreement means that the issue is not as "hotly contested" as Exchange plaintiffs suggest, but does not render evidence directed to that fact irrelevant. See Old Chief v. United States , 519 U.S. 172 , 179, 117 S.Ct. 644 , 136 L.Ed.2d 574 (1997) ("The fact to which the evidence is directed need not be in dispute." (quoting Fed. R. Evid. 401 advisory committee's note)). Indeed, Rabobank's argument here is somewhat circular in that it relies on Dr. Netz's opinion, which it seeks to have excluded, to establish agreement on the point. The trimming methodology applied to LIBOR submissions in order to calculate published LIBOR introduces sufficient complexity such that Dr. Netz's analysis is at least somewhat helpful.

We also reject Rabobank's argument that Dr. Netz's opinion is unreliable because her median-replacement methodology relies on unsupported assumptions. As Dr. Netz disclosed in her initial report, she "do[es] not claim that the median LIBOR submission is necessarily the correct proxy for the but-for LIBOR submissions or that the LIBORs in Exhibit 2 are accurate measures of but-for LIBOR." (Netz Initial Report 29.) Rather, Dr. Netz opines that Rabobank's submissions could have impacted published LIBOR, which her median-replacement methodology is in fact capable of demonstrating. 41 In the absence of any suggestion that she has implemented her methodology in an unreliable manner, we conclude that Dr. Netz's methodology is sufficiently reliable to support her opinion that Rabobank's submissions could impact published LIBOR.

*493 1.2.3. Effect of Changes in LIBOR on EDF Prices

Dr. Netz, like Dr. Seyhun, opines that changes in LIBOR would have affected the prices of EDF contracts and options in a determinable way, basing this opinion on an arbitrage theory, her interpretation of several communications between traders and LIBOR submitters, and a correlation analysis. (Netz Initial Report 32-38.)

Rabobank contends at the threshold that Dr. Netz is unqualified to opine on the relationship between LIBOR and EDFs because her experience in economics broadly does not extend to the specific subject of EDFs and EDF trading. (Rabobank Netz Mem. 13-15.) Exchange plaintiffs' appeals to Dr. Netz's extensive experience in antitrust cases more broadly are not persuasive because the substance of the arbitrage theory that Dr. Netz propounds hardly resembles the issues of "supply, demand, pricing, etc." at play in a more typical antitrust case (Exch. Pls.' Netz Opp'n 15); indeed, no antitrust claims remain in the Exchange plaintiffs' action at all. Nonetheless, we conclude that Dr. Netz's prior research into arbitrage-albeit in physical commodities markets-is sufficient to allow her to opine on the subject in her reports. Indeed, as an economist, Dr. Netz has sufficient training and experience in econometrics in order to conduct statistical testing of correlation and causation and to interpret the results of those tests. 42

Turning to reliability, we consider the bases for Dr. Netz's causation opinion: interpretations of certain trader communications, an analysis of correlations between published LIBOR and EDF prices, Dr. Seyhun's analyses, and two academic papers analyzing the relationship between LIBOR and EDF prices.

First, Dr. Netz interprets several communications between traders and LIBOR submitters to opine that "traders and submitters understood that any changes to LIBOR would immediately and directly impact Eurodollar futures prices." (Netz Initial Report 36.) These opinions impermissibly go to the beliefs and states of mind of those traders and submitters, and are inadmissible. 43 See Highland Capital , 379 F.Supp.2d at 469-70 ; In re Rezulin , 309 F.Supp.2d at 547 .

Second, Dr. Netz calculates the correlations between EDF prices and various tenors of LIBOR. (Netz Initial Report 36-37, ex. 3). From these correlations, her review of Dr. Seyhun's report, and two academic articles, Dr. Netz concludes that changes in LIBOR cause calculable changes in EDF prices. While correlations may be probative of causation in other circumstances, we conclude that they are not in this case: any correlation is confounded by the presence of macroeconomic events. See supra section III.1.1.2.1. Exchange *494 plaintiffs argue that the presence of other factors impacting EDF prices "does not mean that LIBOR's effect disappears in the wake of these other factors," (Exch. Pls.' Netz Opp'n 20), but Exchange plaintiffs have not sufficiently established that LIBOR has an effect as an initial matter and they offer no means of separating the effect of changes in LIBOR (if any) from that of other factors.

Weaker still is Exchange plaintiffs' argument that parsing out "macroeconomic factors affecting interest rates" is unnecessary because they "exist in both the real world and in the but-for world." (Exch. Pls.' Netz Opp'n 20.) As Dr. Netz and Dr. Seyhun both concede, those factors confound a relationship between LIBOR and EDF prices, and the failure to parse them out in assessing the difference between the real world and the but-for world is therefore likely to construct a but-for world that differs from the actual world in ways other than the absence of alleged LIBOR manipulation .

Dr. Seyhun's causation opinions do not, as Exchange plaintiffs suggest, bolster Dr. Netz's causation opinions. (Exch. Pls.' Netz Opp'n 20.) To be sure, an expert may rely on the admissible opinions of another expert, see Gussack Realty Co. v. Xerox Corp. , 224 F.3d 85 , 94-95 (2d Cir. 2000) ; see also Fed. R. Evid. 703, but Dr. Netz is not entitled to rely on Dr. Seyhun's unreliable opinions for support. Indeed, Dr. Netz's opinions suffer from the same tick-size flaw in Dr. Seyhun's opinions: she similarly opines that she is able to determine the probability that EDF contract prices were impacted by changes in LIBOR, an opinion that is unhelpful. 44 See supra section III.1.1.2.3.

Nor do the two academic articles cited by Dr. Netz 45 support her causation opinion. We are dubious that Grinblatt and Jegadeesh's analysis of weekly changes in LIBOR and EDF prices supports Dr. Netz's opinions regarding daily causation given the specific timing issues discussed at length in Dr. Seyhun's and Dr. Culp's reports. See Grinblatt & Jegadeesh at 1516-19. Similarly, Fung and Leung's analysis examines actual Eurodollar rates observed in the market, not LIBOR, see Fung & Leung at 119 n.6 (referring to Eurodollar rate quotes corresponding to the end of the trading day in London), which has different implications for the viability of any arbitrage strategy and, by extension, the relationship between changes in LIBOR and changes in EDF prices ( e.g. , Culp Initial Report ¶¶ 353-61, Culp Rebuttal Report ¶ 26).

Absent further support, Dr. Netz's correlation analysis can show only that-correlation. Because Dr. Netz fails to bridge the analytical gulf between correlation and causation, her opinions that changes in LIBOR caused changes in EDF prices are excluded.

*495 1.2.4. Damages

Finally, Dr. Netz offers several opinions regarding the calculation of damages: (1) that published LIBOR being lower than but-for LIBOR harmed class members who initiated or closed certain positions during different periods (Netz Initial Report 38-44); (2) that all class members were harmed because it is "statistically improbable" that EDF traders are unlikely to have "benefited from or been unharmed by every single act of Defendants' LIBOR manipulation" (Netz Initial Report 44-46); (3) that class members also suffered common harm in the form of the time-value of money in their margin accounts (Netz Initial Report 46-47), and that (4) damages can be calculated on a formulaic basis (Netz Initial Report 48).

As a general matter, these opinions rely on Dr. Netz's earlier opinion that changes in LIBOR cause determinable changes in EDF prices and Dr. Seyhun's opinions to the same effect. Because we determine those opinions to be unreliable, Dr. Netz's derivative damages opinions are therefore insufficiently supported and are therefore inadmissible as well.

We also make three further specific observations. First, the parties' dispute over Dr. Netz's opinion that almost all traders were "harmed" as a result of LIBOR manipulation is, in essence, a dispute over the proper interpretation of the term "harmed." Exchange plaintiffs and Dr. Netz interpret "harmed" to mean "have been negatively impacted at least once" ( i.e. , the trader paid too much money or received not enough money from a single EDF transaction), even if the trader benefited overall from alleged LIBOR manipulation. (Exch. Pls.' Netz Opp'n 22-23). Rabobank interprets "harmed" to mean a negative impact overall once all impacts of LIBOR manipulation have been taken into account. (Rabobank Netz Mem. 20-21). Given our conclusion that Dr. Netz's damages opinions are unreliable, we need not definitively resolve this question here. 46

Second, Exchange plaintiffs' reliance on Mr. Miller's opinions to support Dr. Netz's statement that "data could be obtained from traders themselves, their brokers (FCMs), or the Chicago Mercantile Exchange" is unavailing. (Exch. Pls.' Netz Opp'n 24-25.) Dr. Netz did not rely on Mr. Miller's opinions and instead based her opinion on her "understanding that Eurodollar futures and options are transparent, public markets in which common practice and exchange and federal requirement[s] provide that customers for each transaction be identifiable." (Netz Initial Report 47 n.172.) We question whether Dr. Netz's expertise extends to a subject like the availability of trading records, and Mr. Miller's opinions provide no support because they are inadmissible as we discuss below.

Third, we do not share Rabobank's certainty that Dr. Netz's opinions regarding time-value damages are entirely irrelevant. Compare Helena Assocs., LLC v. EFCO Corp. , No. 06 Civ. 861 (PKL), 2009 WL 2355811 , at *4 (S.D.N.Y. July 29, 2009) (allowing a plaintiff to claim additional damages that were "more properly considered an extension of [its] original damages theory rather than a new theory of liability"), with Point Prods. A.G. v. Sony Music Entm't, Inc. , No. 93 Civ. 4001 (NRB), 2002 WL 31856951 , at *5 (S.D.N.Y. Dec. 19, 2002) (precluding a plaintiff "from changing its damage theory nine years into this litigation"). To the extent these damages may be taken into account, Dr. Netz's *496 opinions regarding time-value damages on the one hand and formulaic calculation on the other are in considerable tension: Dr. Netz does not opine that all EDF traders were subject to the same margin requirements, or that they all shared the same time-value of money. (Netz Initial Report 46-47.) Indeed, an assessment of each trader's margin requirements and personal discount rates appears to necessitate individualized inquiry and does not appear susceptible to formulaic calculation.

1.2.5. Conclusion

Rabobank's motion to exclude Dr. Netz's opinions is granted in part and denied in part. Dr. Netz's opinions derived from her review of trader communications are excluded, regardless of whether she opines that manipulation actually occurred or whether she opines that documents establishing manipulation are available. Dr. Netz's opinions that changes in LIBOR cause calculable changes in EDF prices are excluded based on the reasons discussed at length in our consideration of Dr. Seyhun's opinions, and her opinions regarding damages are excluded as well because they are reliant on her causation opinions. Dr. Netz's opinions regarding the impact of LIBOR submissions on published LIBOR, which address facts in issue, remain admissible.

1.3. Mr. Beevers

Exchange plaintiffs submit two reports from Mr. Craig Beevers: (1) an initial report dated February 2, 2017 (Decl. of David Kovel ex. A, July 10, 2017, ECF No. 2071); and (2) a rebuttal report dated May 3, 2017 (Decl. of David Kovel ex. B, July 10, 2017, ECF No. 2071). We refer to these as the Beevers Initial Report and the Beevers Rebuttal Report. In these two reports, Mr. Beevers offers several opinions across a number of subjects: (1) the adequacy of Rabobank's data productions; (2) a methodology for identifying trader-based manipulation; (3) a Monte Carlo analysis suggesting that panel banks engaged in collusive behavior; and (4) the frequency of alleged manipulation. We conclude that Mr. Beevers is generally qualified to offer these opinions based on his extensive experience in the financial industry. 47

1.3.1. Data Adequacy

Mr. Beevers first offers a number of opinions regarding Rabobank's data productions. After providing an overview of trade reporting and risk reporting systems both generally and at Rabobank specifically (Beevers Initial Report ¶¶ 24-39), Mr. Beevers opines that Rabobank's data productions have been deficient in certain ways, (Beevers Initial Report ¶¶ 46-63), and that trader-level records should be produced because they are necessary to his methodology for identifying trader-based manipulation and the burden to defendants is low. (Beevers Initial Report ¶¶ 40-45, 64-72).

1.3.1.1. Discussion of Trade Reporting and Risk Reporting Systems

Mr. Beevers may address trade reporting and risk reporting in the financial industry generally. We agree with Exchange plaintiffs that his experience provides him with the requisite knowledge and qualifies him to so opine. He may not, however, interpolate these general opinions regarding the financial industry as a whole to Rabobank specifically. Mr. Beevers lacks knowledge as to what data Rabobank in fact possesses, a point that he *497 concedes and Exchange plaintiffs cannot seriously dispute. (Beevers Dep. 408:3-4, Decl. of Robert Lindholm ex. 73, July 7, 2017, ECF No. 2061). While we would view these Rabobank-specific opinions differently had they been anchored in specific legal or regulatory requirements, the fact that other banks possess certain data as a general matter says little about either (1) whether Rabobank ever had this data or (2) Rabobank has maintained this data until now. Therefore, Mr. Beevers's opinions are excluded to the extent they suggest that Rabobank in fact maintains certain reporting systems or possesses certain data.

1.3.1.2. Specific Data Deficiencies

Mr. Beevers next opines that Rabobank's productions are deficient in four specific ways: (1) certain risk management reports are missing; (2) data files contain inconsistent formatting for date information; (3) inconsistent mapping information was provided; and (4) certain datasets appear to be incomplete and contain duplicates.

Mr. Beevers's opinion that certain risk management reports are missing is not admissible for the reasons discussed above: Mr. Beevers lacks knowledge as to what data Rabobank in fact possesses (or possessed) and is forced to speculate as to the existence of those reports. This speculation is plain on the face of his opinions, suggesting, for example, that certain reports "are routinely emailed to desk-wide or division-wide distribution lists on a periodic basis," and that he therefore "believe[s] such reports may exist in Rabobank's email archive." (Beevers Initial Report ¶¶ 46-47 (emphasis added).) We find similarly speculative Mr. Beevers's contention that Rabobank did not produce all of its trade data because the number of trades appears "to be too low by a factor of approximately two" based solely on the opinion of unidentified "experienced money market traders." (Beevers Initial Report ¶ 58.)

As to Mr. Beevers's remaining criticisms, each relies on the implicit assumption that Rabobank possessed immediately tractable data and should have produced it. Mr. Beevers identifies no basis for this assumption, and we find no authority for such a proposition. 48 Rather, the process of data cleansing and ensuring that data are tractable is not only a topic addressed extensively in the literature, see, e.g. , Jonathan I. Maletic & Andrian Marcus, Data Cleansing: A Prelude to Knowledge Discovery , in Data Mining and Knowledge Discovery Handbook , (Oded Maimon & Lior Rokach, eds. 2005) ("Data entry and acquisition is inherently prone to errors, both simple and complex."), but is also a task performed by experts with regularity (including by Dr. Willig in this case (Willig Initial Report app. 4), see, e.g. , DL v. District of Columbia , 194 F.Supp.3d 30 , 40 (D.D.C. 2016) ("Although the primary focus of [the expert's] testimony in those cases has been statistical analysis, he would not have been able to perform that analysis if he had not initially performed the programming and required data cleaning."); Callahan v. City of Chicago , 78 F.Supp.3d 791 , 815 (N.D. Ill. 2015) ("[The expert] 'cleaned' the data set before using it...."); In re Polyurethane Foam Antitrust Litig. , 314 F.R.D. 226 , 278 (N.D. Ohio 2014) (referencing an expert's discussion of "the extent of each Defendant's *498 transactional data and the 'cleaning' steps [the expert] took to make the data usable"). Indeed, Mr. Beevers's expectation appears particularly unrealistic here given the breadth of the data that he seeks and Rabobank's scale as an entity.

We therefore conclude that Mr. Beevers's opinions purporting to identify specific deficiencies in Rabobank's data productions are speculative, and therefore inadmissible.

1.3.1.3. Production

Finally, Mr. Beevers offers a number of opinions addressing whether certain trader-level data must be produced. He opines that trader-level records are "required" in order "to analyze the impact of trader-based manipulation," (Beevers Initial Report ¶ 64), and that this necessity outweighs the burden on defendants of making this production, (Beevers Initial Report ¶¶ 40-41). Therefore, Mr. Beevers concludes, "data quality issues presented by Rabobank not only limit the depth and quality of analysis available to civil litigants, but likely substantially impeded regulatory investigations" and that this poor quality "dictates that Defendants should be required to remediate and supplement the productions to date." (Beevers Initial Report ¶¶ 41, 72.) These opinions are inadmissible both because they are irrelevant and because they offer legal conclusions (including on issues that we have already considered and adjudicated).

These opinions are irrelevant because Exchange plaintiffs need not identify all instances of trader-based manipulation at this stage. Exchange plaintiffs' defense of their relevance is incoherent: they simultaneously contend that Mr. Beevers must explain why he "has not thus far been able to identify all instances of trader-based manipulation at Rabobank during Period 0" on the one hand (Exch. Pls.' Beevers Opp'n 8), and that "Plaintiffs (through Mr. Beevers) are not required to identify all instances of trader-based manipulation at this stage" on the other (Exch. Pls.' Beevers Opp'n 11). Of course, if all instances of manipulation need not be identified at class certification, Mr. Beevers need not explain his failure to do so. Indeed, we highlight here Exchange plaintiffs' assertion that "Mr. Beevers takes no position on" the question of whether "production of Defendants' trade data was necessary for purposes of Plaintiffs' class certification motion." (Exch. Pls.' Beevers Opp'n 11 (emphasis omitted).) Accepting this concession, Mr. Beevers's opinions are plainly irrelevant.

Further, these opinions amount to legal conclusions regarding the scope of discovery, one that impinges on our responsibility to assess whether the discovery being sought is "proportional to the needs of the case." Fed. R. Civ. P. 26(b)(1). To the extent Mr. Beevers's opinions regarding additional production overlap with Exchange plaintiffs' prior motion to compel the production of certain trading records (Letter from Christopher Lovell & David Kovel to the Court, Nov. 4, 2016, ECF No. 1636), those issues have already been resolved by our order denying that motion, see Dec. 6, 2016 Order, ECF No. 1667, and Mr. Beevers's opinions are squarely precluded by that order. To the extent that Exchange plaintiffs rely on Mr. Beevers's opinions to request reconsideration of our order denying Exchange plaintiffs' motion to compel certain trading data, we reaffirm our conclusion that the burden of making available Exchange plaintiffs' requested production-data sufficient to establish trader positions for every single trader at Rabobank over the class period-is not proportional to the relevance and importance of that data, see Fed. R. Civ. P. 26(b)(1). 49

*499 But even if these opinions extend to data not previously sought as Exchange plaintiffs suggest (Exch. Pls.' Beevers Opp'n 11), that breadth does not change their fundamentally legal nature or their inadmissibility. To the extent that they rely on Mr. Beevers's opinions to so argue, we conclude that such a request is remarkably untimely. The class certification schedule for this case was established in February 2016, see Scheduling Order, Feb. 23, 2016, ECF No. 1327, and Exchange plaintiffs offer no explanation for why this information could not have been requested earlier. Rather, they have had more than ample time to seek additional discovery necessary for class-certification purposes, and cannot now complain of any insufficiency. 50

Ultimately, this initial portion of Mr. Beevers's opinions reduces to the proposition that the data Rabobank has produced is not as tractable as he and Exchange plaintiffs would desire for class certification purposes. But Mr. Beevers is not permitted to offer a definition of completeness that is untethered from any concrete requirements (beyond his personal definition of what is "necessary") and then opine that Rabobank's productions fail to meet that wholly subjective standard. Even if these opinions are correct-and the record strongly suggests that at least some are not-they are irrelevant to our consideration of the Exchange plaintiffs' class certification motion. To the extent Exchange plaintiffs assert, relying on Mr. Beevers's expert opinions, that they were afforded insufficient class certification discovery, we reaffirm our decision to deny Exchange plaintiffs' motion to compel and further conclude that the time to litigate class-certification discovery issues has long ago passed.

1.3.2. Identification of Trader-Based Manipulation

Next, Mr. Beevers offers a process for identifying trader-based manipulation, consisting of three different techniques: (1) identifying "anomalous" submissions based on certain mathematical criteria; (2) interpreting trader communications for "corroborating evidence" of manipulation; and (3) analyzing trader positions to see if LIBOR manipulation benefited particular traders. (Beevers Initial Report ¶¶ 73-87; Beevers Rebuttal Report ¶¶ 40-62.) Rabobank and Exchange plaintiffs dispute whether these three components of Mr. Beevers's methodology should be analyzed individually or together; Exchange plaintiffs contend that "Rabobank improperly breaks up Mr. Beevers' methodology into three steps" and suggest that the three components should not be analyzed separately. (Exch. Pls.' Beevers Opp'n 12.)

Mr. Beevers apparently disagrees with Exchange plaintiffs' argument. In responding to Dr. Hubbard's criticism that his methodology is dependent on review of trader communications, Mr. Beevers asserts that his "anomalous submission" criteria are sufficient to identify trader-based manipulation standing alone, as the subsequent steps merely "corroborate" the initial findings of anomalies. (Beevers Rebuttal Report ¶ 57.) But even if Mr. Beevers's methodology were to consist of three steps to be considered together, Daubert requires as a matter of law that "an expert's analysis be reliable at every step." Amorgianos , 303 F.3d at 267 . Because "any step that renders the analysis unreliable under *500 the Daubert factors renders the expert's testimony inadmissible," id. (emphasis omitted) (quoting In re Paoli , 35 F.3d at 745 ), we analyze Mr. Beevers's three techniques individually.

1.3.2.1. Anomalous Submissions

First, Mr. Beevers offers a two-part definition for what constitutes an "anomalous" submission. Under the first part, for a given bank and given tenor, a submission is considered anomalously high (or low) if three conditions are met: (1) the bank made a submission greater (or less) than published LIBOR; (2) the bank's submission is higher (or lower) than the submission made the prior day; and (3) compared to the prior day, the increase (or decrease) in the bank's submission was greater in magnitude than the corresponding change in published LIBOR. (Beevers Rebuttal Report ¶ 47.) Under the second part, for a given bank and any two tenors, one or both submissions 51 are considered anomalous if two conditions are satisfied: (1) the bank increased its submission in one tenor compared to the day before and decreased the submission in the other tenor, and (2) such directionally dissimilar changes are observed for fewer than 25% of panel banks on the day in question. (Beevers Rebuttal Report ¶ 50.)

We agree with Exchange plaintiffs to the limited extent they contend that an expert opinion need not have been peer reviewed and need not be generally accepted in the community in order to be admissible. See Daubert , 509 U.S. at 593 , 113 S.Ct. 2786 (holding that "peer review" is "relevant, though not dispositive" and rejecting the general acceptance standard set forth in Frye v. United States , 293 F. 1013 (D.C. Cir. 1923) ). We agree with Rabobank, however, that Mr. Beevers's methodology for identifying anomalous submissions is insufficiently reliable.

As an initial matter, Mr. Beevers provides scant explanation for how this two-part definition for anomalous submissions was developed. Mr. Beevers's initial report simply defines "anomalous" in a footnote without further explanation. (Beevers Initial Report ¶ 76 n.7.) Standing alone, this definition would be little more than Mr. Beevers's inadmissible ipse dixit that the submissions in question were anomalous. Mr. Beevers's rebuttal report is slightly more detailed, explaining that his criteria are designed to capture instances where one bank "has raised its submission from one day to the next by an inexplicably large amount" and where changes across tenors cannot be explained by "a general [yield] curve steepening or flattening," but the jump from this explanation to the specific mathematical criteria applied appears to be supported only by his ipse dixit , which is insufficient. (Beevers Rebuttal Report ¶¶ 45-54, n.16.)

Indeed, it appears that the first part of the two-part definition would tend to find, for banks whose submissions were consistently above (or below) published LIBOR, anomalous submissions over periods when interest rates were consistently increasing (or decreasing), as the first two criteria would generally be satisfied. As to the second, Mr. Beevers does not explain whether one or both submissions are considered anomalous if the two criteria are satisfied. Accepting counsel's representation at oral argument that the determination of whether one or both submissions *501 are considered anomalous depends on further inquiry into the circumstances (Hr'g Tr. 16:15-19), Mr. Beevers's reports do not explain when both submissions are considered anomalous, when only one submission is considered anomalous, and if only one submission, which one of the two. Absent some explanation as to how these additional "circumstances" and "facts" are analyzed, and given the implementation error that Mr. Beevers has made in applying these criteria (Hubbard Rebuttal Report ¶ 53 n.111), we cannot conclude this part of Mr. Beevers's methodology is sufficiently reliable such that it is admissible.

Our hesitation is confirmed by the nonsensical results produced by these criteria when they are applied to actual LIBOR submissions made outside of Period 0 and the but-for LIBOR submissions calculated by Dr. Seyhun. Specifically, Mr. Beevers's methodology characterizes as anomalous 27 percent of Rabobank's 3-month LIBOR submissions made between 2000 and 2002, and 19 percent of Rabobank's 3-month LIBOR submissions made in 2003 and 2004, 52 both of which are greater than the 17 percent identified during Period 0. (Hubbard Rebuttal Report ¶ 58, ex. 29.) Exchange plaintiffs do not dispute these statistics, but assert that an evaluation of the results produced by Mr. Beevers's criteria is "impermissible under Daubert ," and that these results are not implausible because some evidence suggests that trader-based manipulation occurred as early as 2000. (Exch. Pls.' Beevers Opp'n 13-14.)

But as we have discussed, methodology and results "are not entirely distinct from one another," Amorgianos , 303 F.3d at 266 (quoting Joiner , 522 U.S. at 146 , 118 S.Ct. 512 ), and indeed, Rabobank's robustness testing of Mr. Beevers's methodology relates directly to two of the reliability factors identified by the Supreme Court: whether a methodology "can be (and has been) tested" and a methodology's "known or potential rate of error," Daubert , 509 U.S. at 593 -94 , 113 S.Ct. 2786 . In this case, the high rates of anomalous submissions between 2000 and 2004 identified by Mr. Beevers's methodology suggests that it was not well-tested and that it has a high potential rate of error.

Exchange plaintiffs additionally respond that Rabobank engaged in trader-based manipulation as early as 2003 and that there is therefore no reason to expect Mr. Beevers's methodology to show zero instances of anomalous submissions during this time. But even assuming that Rabobank engaged in trader-based manipulation as early as 2003, Exchange plaintiffs offer no explanation for why, considering the three periods 2000-2002, 2003-2004, and Period 0, Mr. Beevers's analysis identifies the lowest rate of anomalous submissions in the third (17%) and the highest rate of anomalous submissions in the first (27%) when Exchange plaintiffs' allegations and proffered evidence of trader-based manipulation is strongest in the third and concededly nonexistent in the first. While sporadic findings of anomalous submissions prior to Period 0 might be considered indicia of only a minor flaw in Mr. Beevers's criteria, the substantial rates of anomalous submissions found prior to Period 0 and the exactly-backwards relationship between Mr. Beevers's findings of manipulation and Exchange plaintiffs' allegations suggest that Mr. Beevers's methodology has a high rate of error and is insufficiently reliable. 53

*502 Similarly, Mr. Beevers's criteria characterize as anomalous a substantial percentage of the but-for LIBOR submissions calculated by Dr. Seyhun-that is, the submissions Dr. Seyhun contends that panel banks would have made in the absence of LIBOR manipulation . 54 Specifically, Mr. Beevers's methodology identifies as anomalous 5% to 35% of Rabobank's "but-for" LIBOR submissions calculated by Dr. Seyhun, depending on which of Dr. Seyhun's models is considered. Indeed, for two of Dr. Seyhun's models, Mr. Beevers's methodology finds more anomalies in Rabobank's but-for LIBOR submissions than in Rabobank's actual (allegedly manipulated) submissions. (Hubbard Rebuttal Report ¶¶ 70-72, ex. 31.)

In response, Exchange plaintiffs assure us that any inconsistency does not bear on the reliability of Mr. Beevers's opinions because (1) the anomalous-submission definition is only the first step of Mr. Beevers's overall methodology and (2) any disagreement between Mr. Beevers and Dr. Seyhun need not be resolved now. (Exch. Pls.' Beevers Opp'n 14-15). Though Exchange plaintiffs accuse Rabobank of making an "illogical, apples-to-oranges comparison" by comparing Mr. Beevers's "anomalous submissions" analysis to the results of Dr. Seyhun's methodology (Exch. Pls.' Beevers Opp'n 15), Mr. Beevers himself asserts that his analysis is not necessarily reliant on subsequent steps, (Beevers Rebuttal Report ¶ 57). And based on our previously expressed skepticism that the concept of a "battle of the experts" can extend to this extent of disagreement between a party's own experts, we conclude that poor performance of Mr. Beevers's anomalous-submission definition when applied to Dr. Seyhun's findings strongly suggests an untested methodology with an unacceptably high rate of error.

1.3.2.2. Trader Communications

Having identified submissions of interest based on his two two-part definition of "anomalous submissions," Mr. Beevers next proposes to review various trader communications, including "emails, instant message conversations, and recorded conversations." (Beevers Rebuttal Report ¶¶ 11, 57.) We conclude this step is unreliable because it amounts to no methodology at all.

Exchange plaintiffs assert that Mr. Beevers has in fact offered a methodology, "part of which involves applying Mr. Beevers' vast experience and knowledge gained through his work in the dealing rooms of financial institutions." (Exch. Pls.' Beevers Opp'n 15.) Such a vague methodology is not a methodology at all, and it is certainly not a methodology that can or has been tested or one with a known or potential rate of error. In the absence of some recognizable, describable methodology beyond an appeal to Mr. Beevers's qualifications, his opinions that certain documents "clearly show" manipulation whereas others do not (Beevers Rebuttal Report ¶ 58) would be "the essence of unverifiable subjectivity, amounting to the sort of ipse dixit connection between methodology and conclusion" that is properly excluded, Nimely , 414 F.3d at 399 . 55

*503 1.3.2.3. Individual Trader Positions

At step three, Mr. Beevers proposes to "consider the positions of individual traders on [each day with an anomalous submission] to see if there was a relationship between the profit motive (trader position such that a trader or group of traders would benefit from the anomalous submission) and the submission." (Beevers Rebuttal Report ¶ 42.) Specifically, Mr. Beevers intends to undertake "a comparison of which traders were long or short on each particular day with unusually high or low LIBOR submissions," with a "high correlation" between trader positions and anomalous LIBOR submissions serving as "strong corroboration for manipulation." (Beevers Rebuttal Report ¶ 55.) Mr. Beevers does not actually undertake such an analysis, asserting that "the lack of reliable trader-level position data limits [his] ability" to conduct it, but opines that doing so would be feasible once "Defendants produce a complete set of accurate historical trading data for each day during the Class Period that identifies the relevant trader for each position." (Beevers Initial Report ¶ 75.)

Rabobank contends that this step of Mr. Beevers's analysis is governed only by arbitrary criteria susceptible to cherry-picking, especially in light of Mr. Beevers's opinions that different traders within a single bank will often have different exposures to LIBOR (and therefore different profit motives) and in light of Mr. Beevers's failure to explain what he means by a "relationship" between traders' profit motive and a bank's LIBOR submission or what constitutes a "high correlation." (Rabobank Beevers Mem. 11-12.) Exchange plaintiffs respond that Mr. Beevers intends to "perform statistical analysis that would detect those traders who systematically benefit from Rabobank's LIBOR submissions over time," namely, significance testing of the relationship between a trader's positions and Rabobank's LIBOR submissions. (Exch. Pls.' Beevers Opp'n 17.)

As an initial matter, we share Rabobank's skepticism as to whether Mr. Beevers's reports can be fairly interpreted to propose statistical testing. (Rabobank Beevers Reply 8.) The closest Mr. Beevers comes to setting forth any statistical test (which he would perform once he receives all the data he desires) is referencing a "high correlation" serving as "strong corroboration for manipulation." (Beevers Rebuttal Report ¶ 55.)

This lack of detail renders unpersuasive plaintiffs' defense of Mr. Beevers's opinions, even accepting plaintiffs' post hoc exegesis of his methodology. Mr. Beevers could have selected certain statistical tests to perform, discussed the data series to which he would have applied his selected tests (even if the series themselves have not yet been generated), identified what his null hypothesis would have been, and determined the level of statistical significance he would have demanded in order to reject the null hypothesis. 56 But he has *504 done none of those things, and his failure to do so makes examining the reliability of his methodology essentially impossible. While incomplete data may, understandably, impair an expert's ability to fully conduct the analyses that he proposes, it does not excuse the failure to construct an analytical framework under which the data is to be analyzed. The flaw in Mr. Beevers's opinion is the latter, not the former, and we therefore conclude that his opinion is inadmissible.

Of course, even if Mr. Beevers's correlation methodology had been more fully developed, we would conclude that any opinions resulting from the application of that methodology would be inadmissible for lack of reliability. The first step of Mr. Beevers's analysis is unreliable for the reasons discussed above, and any correlation analysis between trader positions and Rabobank's LIBOR submissions to be performed at the third step will rely on those "anomalous submissions" identified at that first step as an input. (Beevers Rebuttal Report ¶¶ 42, 55.) Because Mr. Beevers would not be conducting this correlation analysis using reliable data, any results that it produced would be similarly unreliable. See Nimely , 414 F.3d at 397 (requiring both "reliable data and methodology"); see also Fed. R. Evid. 702(a)(2).

In sum, we conclude that Mr. Beevers's proposed three-step method for identifying trader-based communications is inadmissible. Each step is unreliable as we have explained, and three unreliable steps does not an admissible methodology make.

1.3.3. Monte Carlo Analysis

Next, Mr. Beevers conducts a statistical "Monte Carlo" analysis. "Monte Carlo simulation involves repeated random sampling of various potential scenarios to compute an average result [and is] a technique which obtains a probabilistic approximation to the solution of a problem by using statistical sampling techniques." Advanced Analytics, Inc. v. Citigroup Glob. Markets, Inc. , No. 04 Civ. 3531 (LTS)(HBP), 2009 WL 7133660 , at *1 n.5 (S.D.N.Y. Aug. 5, 2009), report and recommendation adopted in part and rejected in part , 2010 WL 4780772 (S.D.N.Y. Nov. 22, 2010). Based on this analysis, Mr. Beevers opines that further merits discovery will allow plaintiffs to discover more instances of trader-based manipulation, as he concludes that "submissions are not independent and are thus consistent with collusion." (Beevers Initial Report ¶¶ 88, 100.)

As with Dr. Seyhun and Dr. Netz's probabilistic opinions, we conclude that Mr. Beevers's opinion here is irrelevant at class certification and not helpful to the trier of fact. Even assuming that Mr. Beevers's Monte Carlo analysis is reliably implemented, it cannot prove that any trader-based manipulation actually occurred, let alone on which days, by which banks, the extent to which it actually impacted LIBOR submissions, or whether those findings pertain to all class members or only certain class members. Indeed, Mr. Beevers's Monte Carlo analysis, by construction, yields results for an aggregated period and across pairs of banks at its most detailed level. Therefore, contrary to Exchange plaintiffs' contentions, Mr. Beevers's Monte Carlo analysis neither "serve[s] to show Rabobank's anomalous submissions" nor "identif[ies] collusive behavior between panel banks" with any specificity. (Exch. Pls.' Beevers Opp'n 19.)

But even if Mr. Beevers's analysis were helpful to the trier of fact, it would founder on reliability grounds. Mr. Beevers's Monte Carlo simulation analyzes the relationship between a pair of banks' LIBOR submissions for every pair of panel banks, and analyzes the probability that the submissions *505 are both above or both below a reference rate (that is, the submissions have the same "directionality" compared to the reference rate). (Beevers Initial Report ¶¶ 88-110.) Mr. Beevers considers two different reference rates: published LIBOR and the mode of the panel banks' LIBOR submissions ( i.e. , the most frequent submission on a given day). 57

Mr. Beevers then calculates, through random simulation, the expected probability that, conditional on one bank's submission having one directionality, that each other bank's submission has the same directionality. In performing this calculation, Mr. Beevers first determines the percentage of time each bank's submission is above, equal to, or below the reference rate. (Beevers Initial Report tbls.1-2.) With these directionality distributions calculated, Mr. Beevers then runs a large number of simulations. 58 In each simulation, using those distributions calculated at the first step, Mr. Beevers randomly determines, for each bank, whether that bank's submission was above, at, or below the reference rate. Mr. Beevers then aggregates the results of these numerous simulations in order to calculate the "expected" probabilities of one bank's submissions having a certain directionality given another bank submissions having the same directionality, under the assumption that the directionality of one bank's LIBOR submission is independent of the directionalities of all other banks' LIBOR submissions.

Mr. Beevers then compares these "expected" directionality probabilities to the actual probabilities observed, and concludes that the deviation of the actual probabilities from the expected probabilities are generally statistically significant. (Beevers Initial Report ¶ 101.) From this result, Mr. Beevers opines that "submissions are not independent and are thus consistent with collusion." (Beevers Initial Report ¶ 100.)

At the outset, we note that randomness is a deliberate feature of Monte Carlo simulation and reject Rabobank's arguments to the extent they suggest that this randomness by itself is a flaw rendering this analysis unreliable. (Rabobank Beevers Mem. 15.) Under Mr. Beevers's framework, each bank's submission on a given day can fall into one of three categories, meaning that for 16 banks, there are three to the 16th power, or more than 43 million, permutations of the relationships of each bank's submission to the reference rate. The "expected" probabilities that Mr. Beevers calculates would, most rigorously, be calculated based on all 43 million permutations. But given the computational difficulties involved in so calculating, Monte Carlo simulation is helpful in arriving at a reasonable approximation for the "expected" probabilities that Mr. Beevers presents.

While Monte Carlo simulation is a generally accepted technique, general acceptance does not end the reliability inquiry. 59 Problems in implementing the technique *506 may nonetheless render it unreliable as applied, Mr. Beevers's Monte Carlo simulation includes at least two flaws undermining its reliability. First, the Monte Carlo simulation relies on probability distributions of each bank's LIBOR submissions being above, equal to, or below, the reference rate calculated based on Period 0; and second, Mr. Beevers's assumption that each bank's directionality must be independent of all other banks' directionalities is not only unsupported, but demonstrably false.

Mr. Beevers does not identify the time period over which he draws his initial distribution of each bank's submissions compared to the reference rate. However, counsel at oral argument clarified that Mr. Beevers conducted his analysis over Period 0 (Hr'g Tr. 17:9-13), and this representation is consistent with Mr. Beevers's use, for almost all banks and tenors analyzed, of 6

This text is long and has been trimmed here. Open the source document for the complete record.