Glickenhaus & Company v. Household International, Inc.

787 F.3d 408, 2015 U.S. App. LEXIS 8424, 2015 WL 2408028

• Court: Court of Appeals for the Seventh Circuit
• Decided: May 21, 2015
• Docket: 13-3532
• Status: Published

Opinion

SYKES, Circuit Judge.

This securities-fraud class action was tried to a jury and produced an enormous judgment for the plaintiffs — $2.46 billion, apparently one of the largest to date. 1 The defendants are Household International, Inc., and three of its top executives. 2 They challenge the judgment on many grounds, but their primary contention is that the plaintiffs failed to prove loss causation. Proving this element takes sophisticated expert testimony, and the plaintiffs hired one of the best in the field.

The defendants broadly attack the expert’s loss-causation model. They also make the more modest claim that his testimony did not adequately address whether firm-specific, nonfraud factors contributed to the collapse in Household’s stock price during the relevant time period. This latter argument has merit, as we explain below.'

The defendants also raise a claim of instructional error under Janus Capital Group, Inc. v. First Derivative Traders, — U.S. —, 131 S.Ct. 2296, 180 L.Ed.2d 166 (2011), which clarified what it means to “make” a false statement in connection with the purchase or sale of a security. This claim too has merit, but only for the three executives and only for some of the false statements found by the jury. Household itself “made” all the false statements, as Janus defined that term.

The remaining challenges fail. A new trial is warranted on these two issues only. We remand for further proceedings consistent with this opinion.

I. Background

This case is complex and has a lengthy procedural history dating to 2002; retracing it would require a tome. To simplify, we’ll start with the view from 10,000 feet and add details relevant to particular issues as needed.

Household’s business centered on consumer lending-mortgages, home-equity loans, auto financing, and credit-card loans. In 1999 company executives implemented an aggressive growth strategy in pursuit of a higher stock price. Over the next two years, the stock price rose dramatically, but the company’s growth was driven by predatory lending practices. This in turn increased the delinquency rate of Household’s loans, which the executives then tried to mask with creative accounting. Their technique was to “re-age” delinquent loans to distort a popular metric that investors use to gauge the quality of loan portfolios: the percentage of loans that are two or more months delinquent. Household also improperly recorded the revenue from four credit-card agreements, though it ultimately issued corrections in August 2002.

Between the summers of 1999 and 2001, Household’s stock rose from around $40 per share to the mid $60s, and by July of 2001 was trading as high as $69. But the reality of Household’s situation eventually caught up with its stock price. The truth came to light over a period of about a year through a series of disclosures that began when California sued Household over its predatory lending. Other states also launched investigations and eventually collaborated in multi-state litigation. The so-called “disclosure period” culminated when Household settled the multi-state litigation for $484 million. Between the filing of California’s suit on November 15, 2001, and the multistate settlement on October 11, 2002, Household’s stock dropped 54%, from $60.90 to $28.20. Comparatively, declines in the S & P 500 and S & P Finan-cials indexes during this -period were 25% and 21%, respectively.

In 2002 the plaintiffs filed this securities-fraud class action under § 10(b) of the Securities Exchange Act of 1934, 15 U.S.C. § 78j(b), and the Securities and Exchange Commission’s Rule 10b-5, 17 C.F.R. § 240.10b-5, alleging that on numerous occasions Household and its executives misrepresented its lending practices, delinquency rates, and earnings from credit-card agreements. The parties stipulated to class certification, and most issues were tried to a jury over a period of more than three weeks. Jurors were given 40 separate statements that the plaintiffs claimed were actionable misrepresentations. For each statement they were asked to deter mine: (1) whether the statement was actionable (that is, was it false or misleading, material, and caused loss); (2) who among the four defendants was liable for it; (3) which of the three bad practices the statement related to; and (4) whether the particular statement was made knowingly or recklessly by each defendant. Of the 40 possibilities, the jury found 17 actionable misrepresentations and answered the remaining questions.

The jury was also asked to determine how much Household’s stock was overpriced due to' the misrepresentations. The plaintiffs’ expert presented two models for measuring stock-price inflation. Each model generated a table that estimated inflation on any given day during the.class period. The jury adopted one of the two models and used the figures from the corresponding table to complete the special verdict. (We will have more to say about the loss-causation models later.)

This concluded Phase I of the proceedings. In Phase II the parties addressed reliance issues and calculated damages for individual class members. In the meantime, the defendants challenged the jury’s verdict in a motion for judgment as a matter of law, or alternatively, for a new trial. See Fed.R.Civ.P. 50(b). The district court denied the motions.

Some individual claims have yet to be resolved, but the district court entered final judgment on claims totaling $2.46 bil- ■ lion, finding no just reason for delay. See Fed. R. Civ. P. 54(b). This appeal followed.

II. Discussion

The basic elements of a Rule 10b-5 claim are familiar. The plaintiffs had to prove “(1) a material misrepresentation or omission by the defendants]; (2) scienter; (3) a connection between the misrepresentation or omission and the purchase or sale of a security; (4) reliance upon the misrepresentation or omission; (5) economic loss; and (6) loss causation.” Halliburton Co. v. Erica P. John Fund, Inc., — U.S. —, 134 S.Ct. 2398, 2407, 189 L.Ed.2d 339 (2014) (internal quotation marks omitted).

The issues on appeal cluster around three elements. First, and most prominently, the defendants attack the evidence of loss causation. This argument has several layers, but in general the defendants claim that the plaintiffs’ evidence of loss causation was legally insufficient, entitling them to judgment as a matter of law, or at the very least a new trial. Second, they argue that the district court incorrectly instructed the jury on what it means to “make” a false statement in violation of Rule 10b-5, also warranting a new trial. Finally, they contend that discovery rulings during the Phase II proceedings deprived them of a meaningful opportunity to prove that class members did not rely on the misrepresentations.

Different standards of review apply. We review de novo the denial of a motion for judgment as a matter of law; we will reverse only if the evidence was legally insufficient for the jury to have found as it did. Venson v. Altamirano, 749 F.3d 641, 646 (7th Cir.2014); Fed.R.Civ.P. 50(a)(1). We review the denial of a motion for a new trial for abuse of discretion. Venson, 749 F.3d at 656. “A new trial is appropriate if the jury’s verdict is against the manifest weight of the evidence or if the trial was in some way unfair to the moving party.” Id. Jury instructions are reviewed de novo to test whether they fairly and accurately stated the law; a new trial is warranted only if an instructional error caused prejudice. Burzlaff v. Thoroughbred Motorsports, Inc., 758 F.3d 841, 846-47 (7th Cir.2014). We review discovery rulings for abuse of discretion. Thermal Design, Inc. v. Am. Soc’y of Heating, Refrigerating & Air-Conditioning Eng’rs, Inc., 755 F.3d 832, 837 (7th Cir.2012).

A. Loss Causation

We begin, as the defendants do, with loss causation. To prove this element of the claim, the plaintiffs had the burden to establish that the price of the securities they purchased was “inflated” — that is, it was higher than it would have been without the false statements — and that it declined once the truth was revealed. See Dura Pharm., Inc. v. Broudo, 544 U.S. 336, 342-44, 125 S.Ct. 1627, 161 L.Ed.2d 577 (2005); Ray v. Citigroup Global Mkts., Inc., 482 F.3d 991, 995 (7th Cir.2007) (“[PJlaintiffs must show both that the defendants’ alleged misrepresentations artificially inflated the price of the stock and that the value of the stock declined once the market learned of the deception.”). A plaintiffs causal losses are measured by the amount the share price was inflated when he bought the stock minus the amount it was inflated when he sold it. See Dura Pharm., 544 U.S. at 342-44, 125 S.Ct. 1627.

It’s very difficult to know exactly how much stock-price inflation a false statement causes because it requires knowing a counterfactual: what the price would have been without the false statement. It’s tempting to think that inflation can be measured by observing what happens to the stock immediately after a false statement is made. But that assumption is often wrong. For example, say the president of a company lies to the public about earnings (“We made $200 million more than we predicted this year!”) and immediately afterward the company’s stock price rises by $10. The new price could be inflated by exactly $10 if in reality the company had merely met expectations and its stock price would have remained the same had the president told the truth. Or the inflation could be less than $10 if, say, the company really only made $100 million more than predicted and the stock price would have risen by only $5 had the president told the truth. And the inflation might be significantly more than $10 if the company had actually made less than predicted and the stock price would have fallen had the truth been known.

Note too that a stock can be inflated even if the price remains the same or declines after a false statement because the price might have fallen even more (e.g., “We only lost $100 million this year,” when actually losses were $200 million). So the movement of a stock price immediately after a false statement often tells us very little about how much inflation the false statement caused.

The best way to determine the impact of a false statement is to observe what happens when the truth is finally disclosed and use that to work backward, on the assumption that the lie’s positive effect on the share price is equal to the additive inverse of the truth’s negative effect. (Put more simply: what goes up, must come down.) The plaintiffs hired an expert to do exactly that kind of financial analysis: Daniel R. Fischel, founder and president of Lexecon, an economics consulting firm, who at the time also was Professor of Law and Business at Northwestern University and Professor Emeritus at the University of Chicago Law School. 3

Fischel prepared two economic models to quantify the impact of the truth on Household’s stock price. The parties call the simpler of the two the “specific disclosure” model. Fischel identified each major disclosure event and then measured the disclosure’s effect on the stock price on that specific day. This process is more difficult than simply observing how much the price declined; part of the decline may have been caused by market or industry trends. To compensate for this, Fischel used regression analysis to generate a model that predicted the movement of Household’s stock based on the movements in the S & P 500 and the S & P Financials Index, an index of S & P 500 companies in the same industry category as Household. The effect of a disclosure event was calculated as the actual return on the day of the disclosure minus the predicted return (using the regression model and the broader market returns that day). The amount of inflation in the stock price on any given day is then just the sum of the effects of all subsequent disclosures of prior false statements.

To illustrate how the specific-disclosure model works, assume that on the day the world discovers that a company misrepresented its earnings, its stock drops 5%. That same day, however, the market dropped 2% (assume further that the model predicts the company will follow the market generally). The effect of the disclosure is then 3%. If the stock was trading at around $100, then for every day prior to the disclosure, the stock was overpriced by $3. If there was a second disclosure that also caused a $3 drop in price, then for every day prior to both disclosures, the stock was overpriced by $6.

In this ease there were a total of 14 separate disclosure events, and the net effect of. these (some actually caused the stock price to rise) was a decline of $7.97 in the stock price. The final result of the expert’s analysis was a large table listing the amount Household’s stock was overpriced on any given day during the class period, the maximum being $7.97.

One problem with the specific-disclosure model is that the information contained in a major disclosure event often leaks out to some market participants before its release. If this happens, the model will understate the truth’s effect on the price and thus the amount that the stock was overpriced before the truth became known. This is so because the specific-disclosure model only measures price changes on the identified disclosure days and not the effect of more gradual exposure of the fraud. For this reason Fischel provided a second model that the parties refer to as the “leakage” model. This model calculates every difference, both positive and negative, between the stock’s predicted returns (using the same regression analysis described above) and the stock’s actual returns during the disclosure period. The total sum of these residual returns is assumed to be the effect of the disclosures. The amount the stock is overpriced on any given day is the sum of all subsequent residual returns.

As with the specific-disclosure model, the expert’s final product using the leakage model was a table listing these amounts. The total sum of the residual returns during the class period was $23.94, so that figure was treated as a ceiling. In other words, if on any given day the sum of subsequent residual returns exceeded this amount (due to the ups and downs of the market), the. number was replaced with $23.94. Although this model accounts for the movement of the market generally, it does not account for company-specific information unrelated to fraud-corrective disclosures (more on this point in a moment).

The most important thing to understand about both models is that they don’t directly measure inflation caused by false statements; instead they measure the value of the truth. The models tell us that value even if no false statement is ever made because investors might not know the truth for reasons other than false statements (say, for example, if earnings deteriorate before the company needs to report them). As soon as a lie is told, however, the inflation caused by the false statement becomes equal to the value of the truth (as measured by the model) because had the statement been truthful, the stock price would have done what it did do once the truth was revealed. 4

The jurors were given the tables from each model listing the amount the stock was overpriced on each day during the class period (June 30, 1999 to October 11, 2002). They were also given a second table that covered the same period but had blank spaces for each day. Their task was to fill in the amount the stock was overpriced due to misrepresentations. To do so, they had to decide two things: (1) when the first actionable misrepresentation occurred (the plaintiffs claimed there were 40); and (2) which model more accurately measured the effect of disclosures (i.e., the value of the truth). On any date prior to the first actionable misrepresentation, the jurors were told to write a zero. On any date subsequent, they were told to copy .the number from the model they chose.

The jury followed these instructions perfectly. Of the 40 possibilities, the jury-found 17 actionable misrepresentations, the first of which occurred on March 23, 2001. As instructed, the jurors found zero stock-price inflation prior to that date. But starting on that date they adopted and applied the leakage model. That model estimated that the stock was overpriced by $23.94 on March 23, 2001, so the jury’s table does too. The jury’s table thus goes from zero inflation on March 22 to $23.94 worth of inflation on March 23.

The defendants argue that this is absurd. After all, there were 40 possible misrepresentations, 17 of which were found actionable, so how could a single false statement have caused the entire $23.94 of inflation? They also note that Household’s stock only went up $3.40 between March 22 and March 23, and the leakage model’s inflation number only changed by $0.67 (remember, the model only adds on the residual return, which is actual return minus predicted return). The defendants intimate that the March 23 statement couldn’t possibly have caused inflation to increase by any more than these amounts. They make similar observations about the 16 other misrepresentations found by the jury, noting that inflation changes only slightly and sometimes even goes down after each one.

These objections rest on a fundamental misconception about the leakage model. Recall that the amount of inflation caused by a false statement is the difference between the stock price after the false statement and what it would have been had the statement reflected the truth. What the model measures is the effect the truth would have had on the price. Since the net sum of price declines due to corrective disclosures under this model was $23.94, the stock was overpriced by that amount prior to those disclosures. As soon as the first false statement was made, that overpricing became fully attributable to the false statement, even if the stock price didn’t change at all, because had the statement been truthful, the price would have gone down by $23.94 — after all, that’s what it did once the truth was fully revealed. Similarly, every subsequent false statement caused the full amount of inflation to remain in the stock price, even if the price didn’t change at all, because had the truth become known, the price would have fallen then.

Another way to think about it is to see that there are two senses of “inflation.” One is “actual inflation” — -just the difference between the stock price and what the price would have been if the truth had been known; this is what the expert’s model measures. The other is “fraud-induced inflation” — the difference between the stock price and what the price would have been if the defendants had spoken truthfully; this is what the jury determined using the model plus its findings regarding false statements. Before the first false statement is made, there is “actual inflation” in the stock price but no “fraud-induced inflation” because although the stock is overpriced, misrepresentations are not the cause. But as soon as the first false statement is made, fraud-induced inflation becomes equal to actual inflation. Thus, fraud-induced inflation can go from zero to a very large number, even if the stock price doesn’t change at all. In fact, the defendants’ own expert acknowledged that inflation — of the fraud-induced type— can increase even if the stock price doesn’t change and that it must be zero before the first false statement.

The defendants argue that the leakage model of loss causation was legally insufficient because the plaintiffs “made no attempt to prove how Household’s stock price became inflated in the first instance,” but, rather, just “assumed that Household’s stock price was artificially inflated on the first day of the Class Period due to unspecified pre-Class Period misrepresentations and omissions.” They note that Fischel’s tables — under both the specific-disclosure model and the leakage model— show the stock as inflated on the very first day of the class period. True, but neither model assumed that Household’s share price was inflated due to misrepresentations. Instead, the models measure what we have called “actual inflation” — inflation due to investors not knowing the truth. Actual inflation could have resulted from prior misrepresentations or just from the company’s fundamentals having deteriorated without investors knowing about it. How the stock became inflated in the first place is irrelevant because each subsequent false statement prevented the price from falling to its true value and therefore caused the price to remain elevated.

The plaintiffs point all this out in their brief. In reply the defendants object that the plaintiffs are vacillating between two separate and legally distinct theories of loss causation. A false statement that prevents a stock price from falling is an “inflation-maintenance theory,” which (they say) requires the plaintiffs to prove how the inflation was introduced into the stock price in the first place. (There is no law to support this proposition.) If, on the other hand, the plaintiffs are using an “inflation-introduction theory,” then the jury couldn’t possibly have attributed the full $23.94 of inflation to the March 23 statement because the model assumed there was inflation before that date. (This argument conflates actual inflation with fraud-induced inflation, as we’ve just explained.)

More fundamentally, theories of “inflation maintenance” and “inflation introduction” are not separate legal categories. Our decision in Schleicher v. Wendt, 618 F.3d 679 (7th Cir.2010), is instructive on this point. There the parties argued about the distinction between misrepresentations that cause a stock price to rise and those that prevent it from falling, as if that distinction had some legal significance in the loss-causation analysis. We explained why it does not:

When an unduly optimistic false statement causes a stock’s price to rise, the price will fall again when the truth comes to light. Likewise when an unduly optimistic statement stops a price from declining (by adding some good news to the mix): once the truth comes out, the price drops to where it would have been had the statement not been made.... But it should be clear that this is just a mirror image of the situation for the same figures in black ink, rather than red.... Whether the numbers are black or red, the fraud lies in an intentionally false or misleading statement, and the loss is realized when the truth turns out to be worse than the statement implied.

Id. at 683-84.

The Eleventh Circuit agrees:

The district court erroneously assumed that simply because confirmatory false statements have no immediate effect on an already inflated stock price in an efficient market, these statements cannot cause harm. But the inflation level need not change for new investors to be injured by a false statement. Fraudulent statements that prevent a stock price from falling can cause harm by prolonging the period during which the stock is traded at inflated prices. We therefore hold that confirmatory information that wrongfully prolongs a period of inflation — even without increasing the level of inflation — may be actionable under the securities laws.

FindWhat Investor Grp. v. FindWhat.com, 658 F.3d 1282, 1314 (11th Cir.2011).

In short, what the plaintiffs had to prove is that the defendants’ false statements caused the stock price to remain higher than it would have been had the statements been truthful. Fischel’s models calculated the effect of the truth, once it was fully revealed, and the jury found that the defendants concealed the truth through false statements. That is enough.

The defendants have two additional arguments that stand on stronger ground, however. First, they argue that the leakage model, which the jury adopted, did not account for firm-specific, nonfraud factors that may have affected the decline in Household’s stock price. That is true; Fischel acknowledged this in his testimony. The model assumes that any changes in Household’s stock price — other than those that can be explained by general market and industry trends — are attributable to the fraud-related disclosures. If during the relevant period there was significant negative information about Household unrelated to these corrective disclosures (and not attributable to market or industry trends), then the model would overstate the effect of the disclosures and in turn of the false statements. Of course, this can. cut both ways. If during the relevant period there was significant positive information about Household, then the model would understate the effect of the disclosures.

Firm-specific, nonfraud factors were not entirely ignored, however. Although the leakage model doesn’t account for their effect, Fischel testified that he looked for company-specific factors during the relevant period and did not find any significant trend of positive or negative information apart from the fraud-related disclosures:

Q. And did you also analyze whether company-specific factors unrelated to the- alleged fraud can explain Household’s stock price decline during [the disclosure period]?

A. Yes, I did. I looked at that carefully-

I noticed that there were a lot of disclosures that had some fraud-related information in it and some other ... part ... [that] dealt with something other [than that which] was fraud related.

There were some ... of those disclosures that had a positive effect, some had a negative effect; but overall it was impossible to conclude that the difference between the true value line and the actual price would have been any different had there been no disclosures about non-fraud-related information during this particular period. Some positive, some negative. They cancel each other out.

The plaintiffs also introduced e-mails and reports from Household executives attributing the entirety of the stock’s decline to the fraud-related disclosures, and the record contains various reports from market analysts primarily focused on this information. In addition, other evidence loosely corroborates the inflation figure produced by the leakage model ($23.94). For example, when Household embarked on its aggressive growth strategy, one executive (Gary Gilmer, a defendant here) suggested that the stock price could increase by “over 22 dollars a share.”

The defendants contend that this was not enough. Because it was the plaintiffs’ burden to prove loss causation, they argue that the leakage model needed to eliminate any flrm-speciflc, nonfraud related factors that might have contributed to the stock’s decline. This argument relies on the Supreme Court’s opinion in Dura Pharmaceuticals.

The precise issue in Dura is unimportant here, so we’ll describe the case only briefly. The Ninth Circuit had held that in order to plead loss causation in a securities-fraud case, plaintiffs need only allege that the share price was inflated when they purchased their stock. The Supreme Court disagreed, holding that plaintiffs must also allege that the stock price declined once the truth was revealed, because if they also sold their stock while it was still inflated, there would be no loss. 544 U.S. at 342, 125 S.Ct. 1627.

In our case the plaintiffs proved that Household’s share price declined after the truth came out, so the problem identified in Dura is not present here. But the Court’s opinion also contains this very important passage:

If the purchaser sells later after the truth makes its way into the marketplace, an initially inflated purchase price might mean a later loss. But that is far from inevitably so. When the purchaser subsequently resells such shares, even at a lower price, that lower price may reflect, not the earlier misrepresentation, but changed economic circumstances, changed investor expectations, new industry-specific or firm-specific facts, conditions, or other events, which taken separately or together account for some or all of that lower price. (The same is true in respect to a claim that a share’s higher price is lower than it would otherwise have been — a claim we do not consider here.) Other things being equal, the longer the time between purchase and sale, the more likely that this is so, i.e., the more likely that other factors caused the loss.

Given the tangle of factors affecting price, the most logic alone permits us to say is that the higher purchase price will sometimes play a role in bringing about a future loss.

Id. at 342-43, 125 S.Ct. 1627 (second emphasis added).

So in order to prove loss causation, plaintiffs in securities-fraud cases need to isolate the extent to which a decline in stock price is due to fraud-related corrective disclosures and not other factors. See Hubbard v. BankAtl. Bancorp, Inc., 688 F.3d 713, 725-26 (11th Cir.2012); Miller v. Asensio & Co., Inc., 364 F.3d 223, 232 (4th Cir.2004).

Fischel’s models controlled for market and industry factors and general trends in the economy — the regression analysis took care of that. But the leakage model, which the jury adopted, didn’t account for the extent to which firm-specific, nonfraud related information may have contributed to the decline in Household’s share price. Fischel testified — albeit in very general terms — that he considered this possibility and ruled it out. The question is whether that’s enough or whether the model itself must fully account for the possibility that firm-specific, nonfraud factors affected the stock price.

On this point the defendants refer us to several cases rejecting leakage models similar or identical to the one used here. Each of these cases, however, is different in an important respect. For example, one case rejected a leakage model where the plaintiff hadn’t identified any mechanism of corrective disclosure. In re Williams Sec. Litig.-WCG Subclass, 558 F.3d 1130, 1137-38 (10th Cir.2009) (“To satisfy the requirements of Dura, ... any theory— even a leakage theory that posits a gradual exposure of the fraud rather than a full and immediate disclosure — will have to show some mechanism for how the truth was revealed.... The inability to point to a single corrective disclosure does not relieve the plaintiff of showing how the truth was revealed; he cannot say, ‘Well, the market must have known.’ ”). Here, however, the plaintiffs identified 14 separate disclosure events, and they also presented evidence that the content of the disclosures was leaking out to the market gradually prior to their release.

The other cases cited by the defendants rejected the leakage model for failing to account for firm-specific, nonfraud-related information that was both clearly identified and significant in proportion to the disclosures. 5 Here, in contrast, Fischel testified that although there were mixed disclosures during the relevant time period — disclosures that contained both fraud-related and nonfraud information — the nonfraud related information wasn’t significantly positive or negative. Unfortunately, his testimony was very general on this point; neither side bothered to develop it. And the defendants haven’t identified any firm-specific, nonfraud related information that could have significantly distorted the model.

To our knowledge, no court has either upheld or rejected the use of a leakage model in circumstances similar to this case — probably because these cases rarely make it to trial. That said, the Supreme Court has generally recognized that the truth can leak out over time. See Dura, 544 U.S. at 342, 125 S.Ct. 1627 (“But if, say, the purchaser sells the shares quickly before the relevant truth begins to leak out, the misrepresentation will not have led to any loss.”) (emphasis added). So have we. See Schleicher, 618 F.3d at 686 (“[TJruth can come out, and affect the market price, in advance of a formal announcement.”). And other circuits have acknowledged the viability of the leakage theory, at least in principle. See, e.g., In re Flag Telecom Holdings, Ltd. Sec. Litig., 574 F.3d 29, 40-41, 40 n. 5 (2d Cir.2009) (noting the “plausibility” of a “leakage” theory but rejecting it in that particular case because the plaintiffs “failed to demonstrate that any of the information [had] ‘leaked’ into the market”).

The defendants argue that to be legally sufficient, any loss-causation model must itself account for, and perfectly exclude, any firm-specific, nonfraud related factors that may have contributed to the decline in a stock price. 6 It may be very difficult,- if not impossible, for any statistical model to do this. See Janet Cooper Alexander, The Value of Bad News in Securities Class Actions, 41 UCLA L.Rev. 1421, 1452, 1469 (1994); Esther Bruegger & Frederick C. Dunbar, Estimating Financial Fraud Damages with Response Coefficients, 35 J. Corp. L. 11, 25 (2009); Frederick C. Dunbar & Arun Sen, Counterfactual Keys to Causation and Damages in Shareholder Class-Action Lawsuits, 2009 Wis. L. Rev. 199, 242 (2009). Accepting the defendants’ position likely would doom the leakage theory as a method of quantifying loss causation. On the other hand, if it’s enough for a loss-causation expert to offer a concluso-ry opinion that no firm-specific, nonfraud' related information affected the stock price during the relevant time period, then it may be far too easy for plaintiffs to evade the loss-causation principles explained in Dura.

There is a middle ground. If the plaintiffs’ expert testifies that no firm-specific, nonfraud related information contributed to the decline in stock price during the relevant time period and explains in nonconclusory terms the basis for this opinion, then it’s reasonable to expect the defendants to shoulder the burden of identifying some significant, firm-specific, non-fraud related information that could have affected the stock price. If they can’t, then the leakage model can go to the jury; if they can, then the burden shifts back to the plaintiffs to account for that specific information or provide a loss-causation model that doesn’t suffer from the same problem, like the specific-disclosure model. 7 One possible way to address the issue is to simply exclude from the model’s cal- culation any days identified by the defendants on which significant, firm-specific, nonfraud related information was released. See Allen Ferrell & Atanu Saha, The Loss Causation Requirement for Rule 10B-5 Causes of Action: The Implications of Dura Pharmaceuticals, Inc. v. Broudo, 63 BUS. LAW. 163, 169 (2007).

Because this case is one of the few to make it to trial on a leakage theory, the process of submitting the loss-causation issue to the jury was understandably ad hoc. 8 In light of Dura, however, we conclude that the evidence at trial did not adequately account for the possibility that firm-specific, nonfraud related information may have affected the decline in Household’s stock price during the relevant time period. As things stand, the record reflects only the expert’s general statement that any such information was insignificant. That’s not enough. A new trial is warranted on the loss-causation issue consistent with the approach we’ve sketched in this opinion.

The defendants have one final argument about loss causation, which they raise for the first time on appeal. We’ll address it anyway since the problem is easily resolved on remand. The plaintiffs’ leakage model calculates the effect of full disclosure of all three of Household’s bad practices: predatory lending, re-aging delinquent loans, and misrepresenting earnings. The first actionable false statement found by the jury, however, only addressed predatory lending.' Based on the assumptions underlying the leakage model, it can’t be the case that the stock price would have fallen fully had this statement reflected the truth; investors would not yet have learned of Household’s re-aging practices or true earnings.

The defendants correctly note this problem, but it happens to only have a minor effect in this case. The second actionable false statement came on March 28, 2001, only three trading days later, and it covered all three bad practices. Had this statement been true, the market would have been fully informed and the stock would have dropped to its true value. The defendants maintain that this problem undermines the entire model: The effect may be modest here, but what if the jury had found a different first false statement and the gap was much larger?

As support for this position, the defendants rely on Comcast Corp. v. Behrend, — U.S. —, 133 S.Ct. 1426, 185 L.Ed.2d 515 (2013), but that case doesn’t require us to wholly reject the leakage model. Com-cast was a class action alleging that a cable-television provider’s pricing violated the Sherman Act in four separate ways. Id. at 1430. The plaintiffs submitted a damages model that computed what the cable services would have cost but for all four categories of antitrust violations. The issue was whether class certification was appropriate. The district court held that only one of the four theories of antitrust impact was capable of class-wide proof, but the court also held that damages could be proved on a class-wide basis via the plaintiffs’ model. Id. at 1431. The Supreme Court reversed because the model did not separate damages by category. In other words, because the class could only proceed on one theory of antitrust impact, the plaintiffs were left with no correspondingly limited class-wide way to prove damages. Id. at 1434-35.

Here, on the other hand, the jury found that Household and its executives lied about all three categories of bad practices. Accordingly, the Comcast principle applies, at most, to the period between the first false statement and the date — just three days later — on which the jury found actionable false statements addressing all three bad practices. 9

There is a simple solution to this problem: instruct the jurors that if the first actionable misrepresentation relates only to one or two of the three categories of fraud, they should find zero inflation in the stock (or some fraction of the model they’ve chosen) until there are actionable misrepresentations addressing all three. This option wasn’t considered below because the defendants never raised this specific objection (they objected to the leakage model more generally), but the point is that the problem doesn’t defeat the expert’s model.

The defendants do not challenge the jury’s misrepresentation findings, so the 17 actionable false statements are fixed; we need only worry about those three trading days. If the plaintiffs can supply evidence that some fraction of their model is a reasonable estimate of the effect of predatory lending alone, then the new jury may consider that number. Otherwise, the jury should be instructed to enter zero inflation for those three days.

B. Janus Error

Next up is a claim of instructional error. The defendants argue that the jury was incorrectly instructed on what it means to “make” a false statement in violation of the securities laws. The relevant part of the instruction was as follows- (with the offending phrase italicized):

To prevail on their 10b-5 claim against any defendant, plaintiffs must prove ...:

(1) the defendant made, approved, or furnished information to be included in a false statement of fact ... during the relevant time period between July 30, 1999 and October 11, 2002....

After the Phase I trial concluded, and while Phase II proceedings were underway, the Supreme Court issued its decision in Janus narrowly construing what it means to “make” a false statement in violation of Rule 10b-5. The specific issue in Janus was whether a mutual fund investment advisor could be held liable for false statements contained in the prospectuses of its client mutual funds. 131 S.Ct. at 2299. The investment advisor in Janus was wholly owned by the company that created its client mutual funds, and there also was some management overlap. Id. Although the advisor had substantially assisted in the preparation of the prospectuses, it argued that it was not the “maker” of the false statements for purposes of Rule 10b-5. The Supreme Court agreed:

For purposes of Rule 10b-5, the maker of a statement is the person or entity with ultimate authority over the statement, including its content and whether and how to communicate it. Without control, a person or entity can merely suggest what to say, not “make” a statement in its own right. One who prepares or publishes a statement on behalf of another is not its maker. And in the ordinary case, attribution within a statement or implicit from surrounding circumstances is strong evidence that a statement was made by — and only by— the party to whom it is attributed. This rule might best be exemplified by the relationship between a speechwriter and a speaker. Even when a speechwriter drafts a speech, the content is entirely within the control of the person who delivers it. And it is the speaker who takes credit — or blame — for what is ultimately said.

Id. at 2302.

In light of Janus, the defendants moved for a new trial, arguing that the “approved or furnished information” language in the jury instruction misstated the law and had the effect of holding some of them liable for false statements that they did not “make,” as the Supreme Court construed that term. The judge denied the motion, reasoning that the Court’s holding applied only to legally independent third parties (like the investment advisor in Janus itself), not corporate insiders like the individual defendants here, all top executives at Household. 10

That was error. Nothing in Janus limits its holding to legally independent third parties. The Court interpreted the language of Rule 10b-5, which makes it “unlawful for any person ... [t]o make any untrue statement of material fact” in connection with the purchase or sale of securities. 17 C.F.R. § 240.10b-5(b). The Court’s interpretation applies generally, not just to corporate outsiders. 11

And there can be little doubt that the instruction used here directly contradicts Janus. The judge instructed the jury that the plaintiffs could prevail on their Rule 10b-5 claim if they proved that the defendant “made, approved, or furnished information to be included in a false statement.” (Emphasis added.) This goes well beyond the narrow interpretation adopted in Janus. See 131 S.Ct. at 2303 (“Adopting the Government’s definition of ‘make’ would ... lead to results inconsistent with our precedent ... [because it] would permit private plaintiffs to sue a person who ‘provides the false or misleading information that another person then puts into the statement.’ ”). The instruction plainly misstated the law.

Still, we must decide whether this error caused the defendants any prejudice. 12 See Jimenez v. City of Chicago, 732 F.3d 710, 717 (7th Cir.2013). The four defendants in this case are William Aldinger, Household’s CEO; David Schoenholz, the CFO; Gilmer, Vice-Chairman and President of Consumer Lending; and Household itself. Of the 17 actionable false statements, 14 were contained in SEC filings or official Household press releases. The remaining three were delivered by the executives: one was a statement by Gilmer to the media; another was a presentation by Aldinger to Goldman Sachs; and the third was a presentation by Schoenholz at Household’s annual “Investor Relations Conference.”

1. Household

The prejudice analysis is easiest for Household, so we’ll start there. The company stipulated that it “made” all statements in its SEC filings and press releases. That leaves only the three false statements delivered by the three executives. Nothing in Janus undid the longstanding rule that “[a] corporation is liable for statements by employees who have apparent authority to make them.” Makor Issues & Rights, Ltd. v. Tellabs, Inc., 513 F.3d 702, 708 (7th Cir.2008) (citing Am. Soc. of Mech. Eng’rs, Inc. v. Hydrolevel Corp., 456 U.S. 556, 568, 102 S.Ct. 1935, 72 L.Ed.2d 330 (1982)); see also Fulton Cnty. Emps. Ret. Sys. v. MGIC Inv. Corp., 675 F.3d 1047, 1051 (7th Cir.2012) (noting that executives speak for themselves and for their organization). The instructional error clearly did not prejudice Household.

2. Aldinger

Aldinger concedes that he “made” all the statements in Household’s SEC filings and in his own presentation to Goldman Sachs. The plaintiffs claim that Al-dinger also agrees that he “made” the statements in the press releases, but we can’t find that concession anywhere in the record.

We’re hesitant to hold as a matter of law that a CEO “makes” all statements contained in a company press release, as that term was narrowly defined in Janus. We haven’t been directed to evidence showing that Aldinger’s signature or name appeared in the press releases in the sense of an attribution. See Janus, 131 S.Ct. at 2302 (“[I]n the ordinary case, attribution within a statement ... is strong evidence that a statement was made by — and only by — the party to whom it is attributed.”); cf. Peterson v. Winston & Strawn LLP, 729 F.3d 750, 752 (7th Cir.2013) (noting that the defendant law firm would probably not be liable for the contents of a circular it helped prepare because it “did not sign the document or warrant the truth of its contents”). Nor does it appear that he actually delivered the statements in the press releases himself — say, for example, by reading them at a press conference. See Janus, 131 S.Ct. at 2302 (“One ‘makes’ a statement by stating it.”). Ab sent either attribution or actual delivery, the Janus inquiry turns on control. Id. at 2303 (“[T]he rule we adopt today [is] that the maker of a statement is the entity with authority over the content of the statement and whether and how to communicate it.”).

As CEO, Aldinger of course had authority over the press releases in the sense that he could have exercised control over their content. But if that were enough to satisfy Janus, then CEOs would be liable for any statements made by their employees acting within the scope of their employment. That wouldn’t square with the Court’s reminder about “the narrow scope that we must give the implied private right of action” under Rule 10b-5. Id. Instead, as we understand Janus, Aldinger must have actually exercised control over the content of the press releases and whether and how they were communicated. That’s an inherently fact-bound inquiry, and it can’t be answered on this record. Accordingly, as to Aldinger’s liability for the press releases, the Janus error was prejudicial, and he is entitled to a new trial.

The error was not prejudicial as to Aldinger’s liability for Gilmer’s false statement to the media, however. The evidence at trial clearly established that Aldinger “made” this statement in the sense meant by Janus. Aldinger drafted the statement in response to growing protests about Household’s predatory lending practices, and he sent it to various executives, including Gilmer, in an e-mail that said, “Attached to this [e-mail] is our media holding statement....” Gilmer simply read the statement verbatim to the media. As the CEO and the actual author of the statement, Aldinger had the “ultimate authority” over its content and whether and how to communicate it, the touchstone of Janus. Id. at 2302.

The defendants contend that the question of prejudice must be considered in light of the jury’s findings on scienter. They note, for example, that the jury found Household and Aldinger responsible for “making” the Gilmer statement knowingly, while Gilmer, who actually delivered it, was found to have made it recklessly. The defendants suggest that this kind of combination is impossible after Janus. We do not see why. Nothing in Janus precludes a single statement from having multiple makers. See In re Pfizer Inc. Sec. Litig., 936 F.Supp.2d 252, 268-69 (S.D.N.Y.2013); City of Pontiac, 875 F.Supp.2d at 374; City of Roseville Emps.’ Ret. Sys. v. EnergySolutions, Inc., 814 F.Supp.2d 395, 417 (S.D.N.Y.2011). And it’s not illogical to conclude that Aldinger, who wrote the statement and instructed Gilmer to deliver it, acted knowingly, while Gilmer, who simply parroted it, was merely reckless as to its falsity.

That leaves the presentation by Schoenholz at the Investor Relations Conference. The plaintiffs argue that Aldinger’s presence in the room, and his participation in a question-and-answer session afterward, demonstrate that he controlled the content of the presentation, and that’s enough to satisfy Janus. We agree that post-Janus, liability for “making” a false statement can be established by inferences drawn from surrounding circumstances. But we can’t say with confidence that Aldinger’s actions at the conference satisfy the Janus standard. They may, but a properly instructed jury might conclude otherwise.

Finally, the plaintiffs argue that the Janus error cannot have prejudiced Aldinger because he was found secondarily hable under § 20(a) of the Securities Exchange Act, which provides that “[e]very person who ... controls any person liable under any provision of this chapter ... shall also be liable jointly and severally with and to the same extent as such controlled person.” 15 U.S.C. § 78t(a). The jury found, for purposes of § 20(a), that Aldinger and Schoenholz were controlling persons with respect to each other and with respect to Household and Gilmer. Because Household issued the press releases and Schoenholz gave the presentation, this means that Aldinger is secondarily liable for their statements.

Even so, Aldinger may have been affected by the jury’s allocation of responsibility for the plaintiffs’ losses. When multiple defendants are found liable, the jury is required to apportion fault between them. Id. § 78u-4(f)(3). The jury allocated 55% responsibility to Household, 20% to Aldinger, 15% to Schoenholz, and 10% to Gilmer. With a proper instruction on what it means to “make” a false statement, the jury might allocate responsibility differently. 13

Accordingly, Aldinger is entitled to a new trial on whether he “made” the false statements in Household’s press releases and in Schoenholz’s presentation at the Investor Relations Conference.

3. Schoenholz

Schoenholz’s situation is almost identical to Aldinger’s. He concedes that he “made” the false statements in the SEC filing and in his own presentation at the Investor Relations Conference. He was not found liable for Gilmer’s statement to the media. That leaves only the press releases and Aldinger’s presentation to Goldman Sachs. For the reasons already discussed, Schoenholz is entitled to a new trial on whether he “made” those particular false statements, as that term was defined in Janus.

4. Gilmer

As for Gilmer, he actually delivered only one of the 17 actionable false statements. The plaintiffs argue that he was also a “maker” of the false statements in the SEC filings and press releases because as a high-ranking officer, he reviewed and approved them. But the same could have been said for the investment advisor in Janus. And as we’ve already explained, Janus can’t be ignored simply because Gil-mer is a corporate insider. So for all but the statement to the media that he himself delivered, the Janus error prejudiced Gil-mer.

* * *

To summarize, Aldinger is entitled to a new trial to determine whether he was the “maker,” in the Janus sense, of the false statements in Household’s press releases and Schoenholz’s presentation. Schoen-holz is entitled to a new trial to determine whether he “made” the false statements in the press releases and in Aldinger’s pres entation to Goldman Sachs. Gilmer is entitled to a new trial to determine whether he “made” any of the actionable false statements beyond the one he personally delivered to the media.

For clarity’s sake, we add that the.defendants may not relitigate whether any of the 17 statements were false or material. The jury’s secondary liability findings also remain undisturbed. Those issues were not challenged on appeal and do not need to be retried. Of course, the plaintiffs likewise can’t relitigate the other 23 statements. The new trial should focus oh whether the three executives “made” the particular statements we’ve identified and whether they did so knowingly or recklessly. The new jury will also have to reallocate responsibility between the four defendants.

C. Reliance

Finally, the defendants argue that the district court’s Phase II rulings deprived them of a meaningful opportunity to rebut the presumption of reliance. Reliance on a misrepresentation (sometimes called “transaction causation”) is an essential element of a Rule 10b-5 action, but the Supreme Court has recognized a rebuttable presumption of reliance for anyone who purchased a security in an efficient market. See Basic Inc. v. Levinson, 485 U.S. 224, 108 S.Ct. 978, 99 L.Ed.2d 194 (1988); see also Halliburton, 134 S.Ct. 2398 (reaffirming Basic, but holding that defendants can rebut the presumption at the class-certification stage). The presumption recognized in Basic is premised on the “fraud on the market” theory, which posits that “in an open and developed securities market, the price of a company’s stock is determined by the available material information regarding the company and its business.... Misleading statements will therefore defraud purchasers of stock even if the purchasers do not directly rely on the misstatements.” Basic, 485 U.S. at 241-42, 108 S.Ct. 978 (internal quotation marks omitted).

To invoke the presumption, the plaintiffs must prove: “(1) that the alleged misrepresentations were publicly known, (2) that they were material, (3) that the stock traded in an efficient market, and (4) that the plaintiff traded the stock between the time the misrepresentations were made and when the truth was revealed.” Halliburton, 134 S.Ct. at 2408. There’s no dispute that these prerequisites were met here.

The Basic presumption is a strong one. The Supreme Court noted that it’s “hard to imagine that there ever is a buyer or seller who does not rely on market integrity. Who would knowingly roll the dice in a crooked crap game?” Basic, 485 U.S. at 246-47, 108 S.Ct. 978 (quoting Schlanger v. Four-Phase Sys. Inc., 555 F.Supp. 535, 538 (S.D.N.Y.1982)). Even so, the presumption can be rebutted by “[a]ny showing that severs the link between the alleged misrepresentation and either the price received (or paid) by the plaintiff, or his decision to trade at a fair market price.” Halliburton, 134 S.Ct. at 2408 (quoting Basic, 485 U.S. at 248, 108 S.Ct. 978).

Basic identified three circumstances in which the presumption might be rebutted. First, if “ ‘market makers’ were privy to the truth,” then the price would not be affected by misrepresentations. Basic, 485 U.S. at 248, 108 S.Ct. 978. Similarly, if news of the truth had “entered the market and dissipated the effects of the misstatements, those who traded ... after the corrective statements would have no direct or indirect connection with the fraud.” Id. at 248-49, 108 S.Ct. 978. (These two methods of rebutting the presumption are sometimes called “truth on the market” defenses.) Finally, defendants may rebut the Basic presumption by showing that individual plaintiffs traded “without relying on the integrity of the market.” Id. at 249, 108 S.Ct. 978. “For example, a plaintiff who believed ... statements were false ... and ... that [the securities were] artificially underpriced, but sold ... [anyway] because of other unrelated concerns, ... could not be said to have relied on the integrity of a price he knew had been manipulated.” Id.

In Halliburton the Supreme Court summarized these examples as follows:

[I]f a defendant could show that the alleged misrepresentation did not, for whatever reason, actually affect the market price, or that a plaintiff would have bought or sold the stock even had he been aware that the stock’s price was tainted by fraud, then the presumption of reliance would not apply.

134 S.Ct. at 2408.

The defendants argue that the district court’s Phase II procedures deprived them of a meaningful opportunity to rebut the presumption for most class members. Resolving this argument requires some additional procedural background.

1. Phase II Procedures

In Phase I the jury addressed all issues that were appropriate for class-wide resolution — e.g., whether any of the 40 possible false statements were actionable misrepresentations, whether they were material, who was liable for which misrepresentations, and how much inflation the actionable misrepresentations caused in the stock price. Phase II addressed the remaining issues — e.g., reliance questions and the calculation of individual class members’ damages.

The defendants wanted to conduct substantial discovery during Phase II in an attempt to rebut the presumption of reliance for each class member. Initially, however, the judge significantly limited the scope of this discovery. He reasoned that the first two methods of rebutting the Basic presumption — either showing that the market was privy to the truth all along or that the truth had entered the market and dissipated the effects of the misstatements — had already been rejected by the jury in Phase I. The only remaining way to rebut the presumption was for the defendants to show that individual plaintiffs bought or sold Household stock without relying on the integrity of the market.

To streamline discovery on that question, the judge required all class members to answer a preliminary interrogatory:

If you had known at the time of your purchase of Household stock that defendants’ false and misleading statements had the effect of inflating the price of Household[’s] stock and thereby caused you to pay more for Household stock than you should have paid, would you have still purchased the stock at the inflated price you paid? YES_NO

If class members answered this question “no,” then it did not matter how or why they purchased Household’s stock (e.g., via a trading algorithm or as part of a hedging strategy) because they bought at the market price on the assumption that there was no fraud cooked into the price. Only if a class member answered “yes” would the defendants be permitted additional discovery. The court thought this protocol would “sensibly resolve the tension between the rebuttable presumption of reliance and the practicalities and purposes behind [class actions].”

The defendants objected and asked the judge to reconsider, arguing that this procedure unreasonably limited their ability to rebut the presumption of reliance. The judge reversed course and allowed Phase II discovery to proceed while the class members returned their answers to the preliminary question. The defendants asked for a period of 120 days to conduct this discovery. The judge granted the request. The defendants then served 98 class members with document requests, interrogatories, and deposition notices. Among other things, these discovery requests sought “all documents that you reviewed or relied upon in making any decision to engage in any transaction with respect to Household securities.”

The plaintiffs objected that the requests were harassing and far too broad, and that much of the information sought was irrelevant. The court agreed that the requests were overly broad and unnecessary, noting that class members “would have to list every issue of the Wall Street Journal that every employee of that particular institution that dealt in trades read on the subway on the way in to work and back.” So the judge limited the defendants to asking class members about any nonpublic information they relied on in deciding whether to buy or sell Household stock. The judge also permitted the defendants to ask about trading strategies and similar matters, provided the questions were specific and not overly burdensome.

As for depositions, the defendants had earlier advised the court that they would need Phase II depositions from only 10 to 15 large institutional investors. They could not explain why they now sought to depose 98 class members, so the judge imposed a limit of 15 depositions.

Following the court’s limiting order, the defendants served revised written discovery on about 100 class members asking about trading strategies, communications with Household, and any nonpublic information they relied on. They also deposed 12 large institutional investors. At the end of the 120-day discovery period, they asked for more time, even though the judge had indicated at the beginning that he was not inclined to grant any extension. The judge denied the request.

When the time period for answering the court’s preliminary question expired, a large number of class members still had not yet responded. The plaintiffs argued they should not be required to answer the question, or alternatively, should be given more time. The judge allowed the plaintiffs to send the question a second time and divided the class members into two groups: class members with claims larger than $250,000 and class members with claims below that amount. Class members with claims larger than $250,000 would be required to answer the court’s question.

For these larger claims, the case would proceed as follows. If a class member answered “no” — that he wouldn’t have bought the stock had he known it was inflated — and discovery had not produced any evidence indicating otherwise, then the class member was entitled to judgment because there was no triable issue as to reliance. If a class member answered “yes,” then reliance would be resolved in a Phase II trial. And if a class member failed to answer the question, then the defendants were entitled to judgment as to that claim.

The second response period yielded the following results: 10,902 claimants answered “no” to the court’s question (these are the claims at issue on this appeal); 14 133 claimants answered “yes”; and 2,476 claimants failed to answer the question. Approximately 30,000 claims remain unresolved. Most of these are claims of class members who failed to answer the court’s question or claims valued at less than $250,000.

2. The Defendants ’ Arguments

The defendants lodge a general objection that the process we’ve just described unreasonably interfered with their ability to rebut the presumption of reliance. For the most part, however, they don’t specify what the court should have done differently-

The defendants’ primary challenge relates to the phrasing of the preliminary question sent to class members. Instead of asking class members whether they would have purchased Household’s stock if they had known that the price was inflated, the defendants say that class members should have been asked whether they would have transacted if they had known that the statements were false. They argue that the - court’s choice of phrasing “impermissibly baked the Basic presumption into [the] question.”

We disagree. The court’s question accurately reflects the Supreme Court’s description of how the Basic presumption can be rebutted. As the Court noted in Halliburton, “if a defendant could show that ... a plaintiff would have bought or sold the stock even had he been aware that the stock’s price was tainted by fraud, then the presumption of reliance would not apply.” 134 S.Ct. at 2408 (emphasis added). Moreover, the defendants’ preferred phrasing would have swept in too many class members. Some investors may have purchased Household’s stock even if they had known the truth behind the defendants’ misrepresentations, but they would not have paid the price they did. These investors would have to answer “yes” to the defendants’ version of the question. But Basic was very clear that the way to rebut the presumption is to show that the investor would have paid the same price:

Any showing that severs the link between the alleged misrepresentation and either the price received (or paid) by the plaintiff, or his decision to trade at a fair market price, will be sufficient to rebut the presumption of reliance.... For example, a plaintiff who believed that Basic’s statements were false ..., and who consequently believed that Basic stock was artificially underpriced, but sold his shares nevertheless ..., could not be said to have relied on the integrity of a price he knew had been manipulated.

Basic, 485 U.S. at 248-49, 108 S.Ct. 978.

The defendants also generally object to the limitations placed on Phase II discovery. This is a nonstarter. The defendants were allowed to serve written discovery on as many class members as they wanted. And in light of the focus of the rebuttal inquiry, it was reasonable to limit the scope of discovery to class members’ trading strategies and any nonpublic information they relied on in deciding to purchase Household stock. As for depositions, the defendants had earlier advised the judge that 10 to 15 would be enough; they were allowed 15.

Finally, the defendants argue that for most class members, a “no” answer to the preliminary question was “dispositive as to whether the presumption could be rebutted.” This is problematic, they say, because the court’s question was essentially meaningless—all class members could see how they needed to respond in order to recover. The question is imperfect, to be sure, but not quite so meaningless as the defendants suggest. Class members were required to answer under penalty of perjury, and a number of them answered “yes.” An even greater number failed to respond; some may have done so knowing they would have to answer “yes.” True, the vast majority answered “no,” but that’s not unexpected given the strength of the Basic presumption. See Basic, 485 U.S. at 246-47, 108 S.Ct. 978; Schleicher, 618 F.3d at 682.

In any event, the preliminary question was not necessarily the end of the inquiry. Class members were entitled to judgment only if they answered “no” and discovery hadn’t turned anything up. The preliminary question was a useful tool for efficiently resolving most claims. As for the rest, discovery allowed the defendants to “attempt to pick off the occasional class member.” Halliburton, 134 S.Ct. at 2412.

Because the proceedings below were neatly divided into two phases, there’s no need to redo anything in Phase II, even though we are remanding for a new trial on certain issues from Phase I. Assuming the plaintiffs can adequately prove loss causation, the district court may rely on the results from Phase II.

III. Conclusion

In sum, the defendants are entitled to a new trial limited to the two issues we’ve identified here: loss causation and whether the three executives “made” certain of the false statements under Janus’s narrow definition of that term. We reject all other claims of error.

Reversed and RemaNded.

1 . See Reuters, HSBC Faces $2.46 Billion Judgment in Securities Fraud Case, N.Y. Times, Oct. 17, 2013, http://www.nytimes.com/2013/10/ 18/busmess/hsbc-is-fined-2-46-billion-in-securities-fraud-case.html.

2 . Household International, Inc., is now known as HSBC Finance Corp. and is owned indirectly by HSBC Holdings pic.

3 . The defendants acknowledge Fischel's prominence in the field. Apparently he's the expert for this kind of financial analysis; the defendants tried to hire him as well, but they were too late. His consulting firm is now known as Compass Lexecon.

4 . This assumes, however, that the only alternative to a false statement is a true statement. If no statement was an alternative, then the model is much less accurate because it measures the effect of the truth, not the effect of silence. We don’t need to worry about this problem here because most of the misrepresentations were made in legally required corporate filings. A few were made to the press or at conferences, but even these were in response to reports that Household's true situation might not be as it appeared. Thus, "no statement” wasn’t really an option.

5 . In Fener v. Operating Engineers Construction Industry & Miscellaneous Pension Fund (Local 66), 579 F.3d 401 (5th Cir.2009), the loss-causation model was Used to determine the effect of a single press release that contained three parts, two of which were unrelated to disclosures of fraud, but the model failed to account for those parts. The Fifth Circuit-"reject[ed] any event study that shows only how a stock reacted to the entire bundle of negative information, rather than examining the evidence linking the culpable disclosure to the stock-price movement.” Id. at 410 (internal quotation marks omitted). In In re REMEC Inc. Securities Litigation, 702 F.Supp.2d 1202, 1274 (S.D.Cal.2010), the court rejected the expert's model because "each of the five identified disclosures, including the three corrective disclosures, contained multiple pieces of company specific information, some negative, some positive, some allegedly fraud related, and some not.” The other cases cited by the defendants are to the same effect. See United States v. Ferguson, 584 F.Supp.2d 447, 453 n. 7 (D.Conn.2008) (“The defendants cited several confounding factors during the 30-day event window [that the model did not account for].”); In re Omnicom Grp., Inc. Sec. Litig., 541 F.Supp.2d 546 (S.D.N.Y.2008) (explaining that a Wall Street loumal article that caused a drop in stock price was either unrelated to disclosures of fraud or was already known to market participants).

6 . The defendants are joined in this argument by the Securities Industry and Finance Markets Association — an advocacy group representing banks, securities firms, and asset managers — as amicus curiae.

7 . Here, of course, the plaintiffs submitted Fis-chel's specific-disclosure model to the jury as an alternative method for quantifying loss causation. But this method might encounter the same problem, if indeed there was some additional negative firm-specific, nonfraud related information on the same day as a specific disclosure.

8 . We intend no criticism of the district judge. To the contrary, he handled this complex and difficult case with thoroughness and care.

9 . So, for example, if the first false statement only addressed one of three categories of fraud and the second statement addressed the other two categories (but not all three), then the model would be accurate after the second statement because at that point had both statements been truthful, the truth would have been fully known and the price would have fallen to the value it did fall to once the truth was disclosed.

10 . As support for this ruling, the judge relied in part on In re Satyam Computer Services Ltd. Securities Litigation, 915 F.Supp.2d 450 (S.D.N.Y.2013), and In re Smith Barney Transfer Agent Litigation, 884 F.Supp.2d 152 (S.D.N.Y.2012), but neither case held that Janus does not apply to corporate insiders. Smith Barney held that corporate executives who sign documents are the "makers” of the statements contained in the documents even though the company has the ultimate authority over the documents. 884 F.Supp.2d at 163-64. And Satyam held that Janus did not overturn the "group pleading doctrine,” 915 F.Supp.2d at 477 n. 16, a pleading rule for alleging scienter that we rejected long before Janus. See Pugh v. Tribune Co., 521 F.3d 686, 693 (7th Cir.2008).

11 . We note that this issue has divided the district courts. Compare, e.g., City of Pontiac Gen. Emps.’ Ret. Sys. v. Lockheed Martin Corp., 875 F.Supp.2d 359, 374 (S.D.N.Y.2012) (“Janus ... addressed only whether third parties can be held liable for statements made by their clients.... [It] has no bearing on how corporate officers who work together in the same entity can be held jointly responsible....”), with Haw. Ironworkers Annuity Trust Fund v. Cole, No. 3:10CV371, 2011 WL 3862206, at *4 (N.D.Ohio Sept. 1, 2011) {“{Janus’s'] interpretation of the verb 'to make’ is an interpretation of the statutory language ... and therefore cannot be ignored simply because the defendants are corporate insiders.”), and In re UBS AG Sec. Litig., No. 07 Civ. 11225(RJS), 2012 WL 4471265, at *10-11 (S.D.N.Y. Sept. 28, 2012), aff'd 752 F.3d 173 (2d Cir.2014) (rejecting an argument that Janus applies only to third parties and not corporate insiders).

12 .Citing Dawson v. New York Life Insurance Co., 135 F.3d 1158 (7th Cir.1998), the defen dants argue that this kind of error is always prejudicial and automatically requires a new trial. Dawson held that when a jury is instructed on multiple theories, one of which is incorrect, "its verdict must be set aside even if the verdict may have been based on a theory on which the jury was properly instructed.'' Id. at 1165. Other cases suggest something similar. See, e.g., Byrd v. Ill. Dept., of Pub. Health, 423 F.3d 696, 709 (7th Cir.2005); Saturday Evening Post Co. v. Rumbleseat Press, Inc., 816 F.2d 1191, 1197 (7th Cir.1987); Simmons, Inc. v. Pinkerton’s, Inc., 762 F.2d 591, 599 n. 3 (7th Cir.1985). This line of cases has been displaced by more recent Supreme Court decisions holding that this kind of error is reviewed for harmlessness, even in a criminal case. See Skilling v. United States, 561 U.S. 358, 414, 130 S.Ct. 2896, 177 L.Ed.2d 619 (2010); Hedgpeth v. Pulido, 555 U.S. 57, 59, 129 S.Ct. 530, 172 L.Ed.2d 388 (2008) (per curiam).

13 . How exactly this would affect Aldinger legally is somewhat complicated. Liability is generally assigned proportionately using the jury’s determination of responsibility, see 15 U.S.C. § 78u-4(f)(2)(B), but a defendant can be jointly and severally liable if he knowingly violated the law, see id. § 78u-4(f)(2)(A). The jury found him liable for one knowing violation (the one we've just described), though it’s not clear whether that makes him jointly and severally liable for all misstatements or only the one he knowingly made. See Regents of Univ. of Cal. v. Credit Suisse First Bos. (USA), Inc., 482 F.3d 372, 404-07 (5th Cir.2007). Nor is it clear how proportional liability and § 20(a) interact. See Laperriere v. Vesta Ins. Grp., Inc., 526 F.3d 715 (11th Cir.2008) (per curiam). We don’t need to resolve these issues because even if Aldinger is jointly and severally liable for all 17 misstatements — either through § 78u-4(£)(2)(A) or § 20(a) or some combination thereof — he is still entitled to seek contribution from the other defendants. See Musick, Peeler & Garrett v. Emp’rs Ins. of Wausau, 508 U.S. 286, 113 S.Ct. 2085, 124 L.Ed.2d 194 (1993). So the size of his share of responsibility matters.

14 . The $2.46 billion judgment — entered pursuant to Rule 54(b) — reflects claims totaling $1.48 billion plus $986 million of prejudgment interest.