dissenting.
Four floor traders at the Chicago Board of Trade executed “freshening” trades in the March 1991 wheat contract. By selling contracts back and forth, these traders moved to the back of the delivery queue. Freshening adds liquidity to the market near contract expiration, when it is much needed, and enables traders to stay in the market longer, which assists hedgers who want to retain opposite positions. The CFTC insists that all trades be done by open outcry on the floor, without prearrangement. 17 C.F.R. § 1.38(a). Believing that a round-robin of trades among four traders at a uniform price could occur only by prior arrangement, the cftc charged them with unlawful trading practices. (The charges include wash sales, in violation of *9397 U.S.C. § 6c(a)(A), and sales not at bona fide prices, in violation of § 6c(a)(B), but these charges depend on the proposition that the trades were non-competitive. I therefore focus on prearrangement.) The contrary view is that the four traders dealt among themselves because they were the only ones willing to take the risk at the end of the contract, that the prices were uniform because no new information about the demand or supply of wheat arrived at the market during the brief periods needed to accomplish the freshening, and that the events are best explained as an example of what in antitrust law would be called oligopolistic interdependence — a situation that does not support a finding of agreement. See JTC Petroleum Corp. v. Piasa Motor Fuels, Inc., 190 F.3d 775, 780 (7th Cir.1999).
Thirteen years ago the oftc concluded that symmetrical freshening trades at a uniform price are enough by themselves to violate federal law. In re Collins, Comm. Fut. L. Rep. ¶ 22,982 (Apr. 4, 1986), ¶ 23,401 (Nov. 26, 1986). But in Stoller v. CFTC, 834 F.2d 262 (2d Cir.1987), the second circuit disapproved this conclusion and held that, unless the Cftc has adopted a rule defining what practices are forbidden, it must conduct a detailed factual inquiry into the genesis of the trades. After Stoller the agency did not adopt such a rule. Although in our case it held a hearing, in the end the Commission found, exactly as in Stoller, that symmetrical freshening trades violate the Act. Today my colleagues approve this conclusion and create a conflict with Stoller. For its part, the cftC relied heavily on the Collins opinions that the second circuit declined to enforce, adding only the puzzling “rev’d on other grounds, Stoller v. cftc, 834 F.2d 262 (2d Cir.1987).” In re Elliott, 1998 cftc Lexis 25 at *20, Comm. Fut. L. Rep. ¶ 27,-243 (Feb. 3,1998). What “other grounds”? That is the last we hear of Stoller from the Commission — which is to say that, like my colleagues, the oftc did not attempt to reconcile its position with the only judicial authority on the subject. (The penultimate paragraph of the majority’s opinion says that Stoller is about wash sales, while this case is about prearrangement. That’s your classic distinction without a difference. Trades come to be called “wash sales” because of prearrangement. The traders’ actual conduct in Stoller and this case appear to be identical. The cftc demonstrated as much by relying heavily on Collins, the very opinion reversed by Stoller. Prearrangement was the central issue in both cases, and my colleagues do not offer any reason to believe that the second circuit would enforce the administrative order, for at the end of the day the cftc disregarded the evidence collected in the hearing and rested on its a priori view about freshening at a uniform price, exactly as in Stoller.)
What evidence does this record contain? In addition to the trades themselves, there are three kinds. First is testimony by the four traders, each of whom denied prearrangement. The alj believed this testimony, and the cfto gave no reason for disbelieving it other than its view (the very position rejected in Stoller) that round-robin trades by themselves demonstrate prearrangement. Second there was testimony by other traders that the questioned transactions were handled by open outcry in the pit, and that other traders could have participated had they wanted to do so (and been willing to take the risk— much more substantial for other traders with small positions in the contract than for the four who sought only to freshen large positions they already held). This testimony by disinterested observers, if believed — and the alj did believe it — is incompatible with prearrangement and demonstrates as well that the supposed dangers of prearrangement were missing. (More on this below.) The cftc did not give a reason for disbelieving this testimony. The third stripe of evidence was supplied by Hugh Rooney, an investigator for the Division of Enforcement, who testified that the trading pattern was so unusual that only prearrangement could explain its details. My colleagues conclude that Rooney’s testimony is junk science of no probative value, and I agree. With Rooney’s *940views disregarded, the only differences between this record and the record in Stoller favor the traders. Yet my colleagues enforce the cftc’s order. They say that the Commission’s “expert” view is entitled to deference. Unless we are to abandon all judicial review of administrative orders, however, we cannot kowtow to a claim of expertise; the claim must be supported by evidence and reasoning.
Ever since Congress began to establish independent agencies in 1887, it has been customary to refer to a commission’s “expertise.” This is a figure of speech, an honorific, rather than a description of commissioners’ backgrounds and skills. It would be more accurate to call commissioners of the Cftc (and other agencies) “specialists.” None of the Commission’s current members is trained in statistical analysis or game theory, which might assist in understanding the dynamics of interactions among a few traders at the end of a contract. None is an expert in industrial organization, which could contribute to understanding oligopolistic interdependence. None is a financial economist. Only one of the four Commissioners who participated in the order under review had any experience in the trading pits, and he dissented from the decision under consideration. (Before joining the cftc, the three members of the majority were a lawyer, a banker, and a manager of an agricultural conservation program.)
Courts accept agencies’ decisions to the extent that Congress has delegated authority to them, not because their decisions may be supported by unspoken reasoning that is too sophisticated to be explained to mere judges. Delegation entails the power to make policy choices. See FCC v. National Citizens Committee for Broadcasting, 436 U.S. 775, 98 S.Ct. 2096, 56 L.Ed.2d 697 (1978); Chevron U.S.A. Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984). What the cftc has done here does not amount to a reasoned policy choice, however. Challenged by Stoller to draw up a rule (or even announce a policy) governing freshening, the Commission kept silent. Instead we have for review a finding of fact that the four traders prearranged their activities. Under the Administrative Procedure Act, the question now is whether that finding is supported by substantial evidence. 5 U.S.C. § 706(2)(E). What evidence might that be, given that all of the testimony other than Rooney’s favored the traders, and Rooney’s is worthless?
The usual way to prove agreement is co-conspirator testimony and physical evidence consistent with private deals off the trading floor. Yet everyone in a position to know the truth testified that there had been no private arrangement. The alj believed this testimony. I recognize that an agency can disagree with an alj’s credibility determination, see Stanley v. Board of Governors, 940 F.2d 267 (7th Cir.1991), but it needs a good reason. Morris v. CFTC, 980 F.2d 1289 (7th Cir.1992). The usual grounds for upsetting a credibility decision, such as internal contradictions in the testimony and incompatibility between oral and documentary evidence, are missing. The Cftc’s only apparent rationale for disbelieving the traders is that their testimony clashes with its a -priori belief that symmetrical trades must have been prearranged. That circular approach is some distance from substantial evidence. Perhaps the cftc had a more acceptable reason, but it did not give one. (The principal reason advanced in the cftc’s appellate brief — that the alj did not make a credibility decision at all, because he applied the word “truthful” rather than the word “credible” to the traders’ testimony — is ludicrous.) Because an alj’s decision to believe testimony is entitled to weight in the substantial-evidence calculus even if the agency disagrees, see Universal Camera Corp. v. NLRB, 340 U.S. 474, 496-97, 71 S.Ct. 456, 95 L.Ed. 456 (1950), the cftc’s lack of a good reason for its disbelief of the traders leaves the alj’s findings intact.
*941Physical evidence is another ordinary way to show agreement. Yet there are no phone records or writings consistent with prior arrangement. One standard form of evidence — so basic that in other cases the CFTC has insisted on it — is evading open outcry on the trading floor. If traders perpetrate hugger-mugger to avoid scrutiny by their peers, and thus to prevent competitive bids from being made, then it is sound to infer that the fix is in and that the “real” price and quantity were set off the floor. In this case, though, it is undisputed that any other trader who wanted to participate could have done so. Stan Komparda and Robert Williams, both veterans of the pits, testified that the trades were made by open outcry and that they could have participated. Three additional traders, William Allen, John Warner, and Daniel Henning, testified that the pattern of round-robin trading in which petitioners engaged was common in the closing month of a contract. My colleagues’ footnote 5, which refers to “a” veteran trader, understates the force of this testimony. But the cftc itself simply ignored it.
Well, if all the testimony other than Rooney’s contradicts the cftc’s position, and if Rooney lacked a foundation for his conclusions, how could the cfto reach the conclusion it did? It gave two hints.
One is that audit trail irregularities occurred. So much is undisputed, but what makes the irregularities (a word the cftc applies to all errors; it does not connote intentional wrongdoing) the basis for an inference that the trades were prearranged? Trading cards frequently are altered, because traders write things down wrong when fast-breaking events distract their attention. The question is not whether errors occurred in the documentation of the trade, but whether these were unusual, and, if so, whether the nature of the departure from the norm implies that misconduct was afoot. The cftc was silent on both of these questions. It did not find (or point to anything in the record that could support a finding) that the irregularities here were abnormal. If the number and kind of irregularities were normal for this type of trading, then there is no basis for an inference that hanky-panky occurred. As for the second question— whether these errors imply prearrangement — the Commission did not explain its thinking. An outsider may be excused for believing that the Commission got it backward. If the four traders prearranged their dealings, then why aren’t the audit trails flawless? It is letter-perfect trading cards, not normal errors, that would point to an existing deal. (Similarly, in antitrust law, it was the rotation of bids by the phases of the moon that signaled the conspiracy among electrical equipment manufacturers. The random noise of a competitive market would have stamped out any such pattern.)
Perhaps the cfto had some reason for thinking that errors, rather than perfection, show prearrangement — but what matters now is that the Commission did not give a reason. Courts do not think that an expert’s bare conclusion is the end of the inquiry. See, e.g., Huey v. United Parcel Service, Inc., 165 F.3d 1084, 1087 (7th Cir.1999) (“An expert who supplies nothing but a bottom line supplies nothing of value to the judicial process”, quoting from Mid-State Fertilizer Co. v. Exchange National Bank, 877 F.2d 1333, 1339 (7th Cir.1989), and citing many other cases); McMahon v. Bunn-O-Matic Corp., 150 F.3d 651, 657-58 (7th Cir.1998); Vollmert v. Wisconsin Department of Transportation, 197 F.3d 293, 298-99 (7th Cir.1999). All we have on the audit-trail irregularity, however, is a bottom line; all of the reasoning, and all of the intermediate facts (such as whether this audit trail was unusual) have been omitted.
The other non-testimonial approach, and the one on which my colleagues principally rely, is the Commission’s belief that the trading reflected “a precision and symmetry not generally found in competitively executed trades.” (cftc Order at 14, quoted by the majority 202 F.3d at 933 and elsewhere.) That is a proposition of fact. Were these trades unusually symmetrical? *942Do competitive markets generate symmetrical trades, and if so what portion of trades have these characteristics? The record is silent on these issues. Rooney asserted that something unusual had happened, not as a statement of historical fact (he did not bother to analyze how trades occur when markets are operating normally) but as a preconception that cross-examination showed to be unfounded. Thus the cornerstone of the cftc’s decision is bereft of evidence. (That petitioners’ trading sequences were not identical to each other, something my colleagues emphasize, id. at 936-37, is beside the point. No rule of law says that the first round of freshening on a given day sets a baseline to which all other rounds must conform. We need to know how a given pattern relates to the market. Some differences are predictable; to repeat a point already made, the mark of prearrangement would be identical sequences, not the normal differences that characterize competitive markets.)
Suppose this issue had been analyzed by a student of markets skilled in statistical methods. Such a person would have gathered information about trading patterns and inquired whether the pattern in this month was unusual. Then the expert would have asked whether the nature of this departure from the norm could best be attributed to prearranged trades or to some other matter — such as the fact that only four traders were active, so they were necessarily dealing only with each other. This happens all the time in discrimination cases. An expert asks not only how many members of a minority group the employer hired, but also whether this number was expected (which depends on the applicant pool) and, if not, whether it was so unusual that an inference may be drawn that race or sex is an explanatory variable. Courts insist that these experts perform careful work and use statistically valid methods. See, e.g., Hazelwood School District v. United States, 433 U.S. 299, 308-09 n. 14, 97 S.Ct. 2736, 53 L.Ed.2d 768 (1977); Hill v. Ross, 183 F.3d 586, 591-92 (7th Cir.1999); Mister v. Illinois Central Gulf R.R., 832 F.2d 1427 (7th Cir.1987).
No statistical analysis appears, in the Cftc’s opinion, however, and a raw bottom line can’t be “substantial evidence.” Our recent opinion in Chicago Board of Trade v. SEC, 187 F.3d 713 (7th Cir.1999), and the D.C. Circuit’s thoughtful opinion in Bechtel v. FCC, 10 F.3d 875 (D.C.Cir.1993), stress that an agency must act like an expert if it wants the judiciary to treat it as one. In both Chicago Board of Trade and Bechtel the court set aside an agency’s order when it failed to undertake the sort of empirical inquiry necessary to support the factual propositions essential to its decision. Just so here.
The Commission’s opinion is problematic on additional grounds. Let me start with the rationale for banning prearranged trades, leading into the question whether this episode fits the rationale.
Wash sales are said to be bad for two reasons: first, they can give the illusion of price movements, which draws victims into the market (and also undermines the value of price signals); second, they can be used to reallocate profit from one trader to another. Reddy v. CFTC, 191 F.3d 109, 115 (2d Cir.1999); Philip McBride Johnson & Thomas Lee Hazen, I Commodities Regulation § 2.05[5] at p. 2-58 (3d ed. 1999). The classic story is one in which schemers trade back and forth with each other at inflated prices. These trades are safe for the schemers because the transactions offset one another (neither party gains or loses a cent) but dangerous to persons not aware of the sham. Outsiders take the higher price as a real one and may buy at the inflated price. When the wash sales stop, the price returns to its original level and the victims lose. The schemers, who have sold securities or contracts to the victims, make off like bandits. See Daniel R. Fischel & David J. Ross, Should the Law Prohibit “Manipulation” in Financial Markets?, 105 Harv. L. Rev. 503 (1991). In our situation, by contrast, prices were stable during the round-robin trading, and no outsiders were gulled. Oddly, the oftc seems to think that the *943lack of price movement shows that something is amiss, but this implies instead that the Commission does not understand how prices are set in competitive markets. In commodities futures markets, prices move only in response to information about supply and demand for the physical commodity. The four traders were changing places in the delivery queue; such trades by definition produce no new information about supply of or demand for wheat and therefore should not affect price. Only prearrangement could have produced price movements without information significant to consumers. The lack of price movement is therefore a mark in the traders’ favor.
The second concern about prearranged trades is reallocation of profits. Suppose the March-July spread is 21$ per bushel, but the traders have received information about grain supplies leading them to believe that the spread will increase, say to 25$. Orders are supposed to be executed in sequence, and whichever customer is first in line should make a substantial profit. If the traders can engage in a private deal, however, one may transact with another at a 21<c spread for this customer’s account (depriving the customer of the profit) and the buyer will resell the contract at a 25$ spread, obtaining the profit for himself (and presumably sharing some with the cooperating floor trader). Out-of-sequence execution and other non-competitive strategies can be a serious problem when, as in this example, it allows traders to enrich themselves at customers’ expense.
But in our case it is agreed that no such thing happened — and not just because all of the witnesses said that the transactions were done by open outcry, so that any departure from the market price would have induced other traders to leap in. The reason no customer was hurt, or even at risk, is that there was no customer, period. The four traders were trading for their personal accounts.
Neither of the two reasons why the cftc believes that prearranged trades can be bad for futures markets is present. Many trades arranged off the trading floor pose no risk of deceit. Stock markets, which used to follow a rule against prearranged trades, have for years allowed deals to be reached off the floor and executed on it; many trades are now handled completely off-exchange. By most accounts the change improved the efficiency of stock markets. Whether the contrary position that the cftc imposes on futures markets is sensible is not, however, a question presented for our consideration. Nonetheless, important interests are at stake. These are the interests served by freshening trades. Following the unanimous view of commentators, my colleagues allow that freshening serves a useful purpose. 202 F.3d at 928 & n. 2. Permitting the practice encourages more traders to stay active later, which improves liquidity for the whole market. Having the ability to freshen near expiration (and thus avoid delivery) makes futures markets more attractive to speculators, who are essential to efficiency because speculators accept the risk that hedging traders want to divest. See Merton H. Miller, Financial Innovations & Market Volatility 200 (1991). Well, if this is so and freshening serves multiple good purposes, and if there is no interest on the other side of the ledger, then the outcome of this case can do nothing but impose needless costs on all participants in futures markets. That was what led Stoller to set aside the Commission’s order: the second circuit thought that by penalizing round-robin trades as if they had been prearranged, all the Commission would do was inflict a random penalty on freshening. That proposition is still true. The cftc thought the symmetry of the transactions proof of arrangement. But most if not all freshening trades will be symmetrical, if the traders do their jobs right. To complain about symmetry is to complain about freshening in the closing weeks of a contract (when few traders remain) — which is contrary to the premise that freshening is lawful.
*944One final comment about the ofto’s reasoning. The Commission thought it significant that the traders were interdependent — which is to say that they held large open positions and thus could accommodate one another. The Commission remarked in passing (and my colleagues’ opinion repeats, also in passing) that something was suspicious because in the initial trade one trader suffered a loss. One trader always initiated by buying the spread, even though all four wanted (in the end) to sell the spread in order to freshen. The traders say that this is normal, that once they see that a round of freshening is in prospect they are willing to buy first because they understand that when the cycle is completed all books will balance. Suspicion then arises: why didn’t the trader who got the gain in the first deal break the circle and keep the gain, leaving the others holding the bag? The answer is: “fool me once, shame on you; fool me twice, shame on me.” The trader who breaks the cycle will be shunned and excluded from freshening in future months, much to his loss.
You don’t need to be an expert in game theory to see this point, but it can’t hurt that formal analysis demonstrates that the prospect of repeat dealing induces people to adopt cooperative behavior without agreement. The prospect of future interactions, and not explicit deals, is what keeps games of mutual interdependence going. See Douglas G. Baird, Robert H. Gertner & Randal C. Picker, Game Theory and the Law 159-87 (1994). Does the cfto really think that it has falsified central conclusions of modern game theory? No, it doesn’t. It has simply ignored the subject.
What to do about the way in which repeat interactions lead small numbers of players to act as if they had agreed is a question that comes up in many parts of the law. Think of oligopolistic interdependence in antitrust. A market has three large firms, with equal shares. None of the three leads the way in cutting prices, even though each knows that it could improve profits in the short run by cutting price and making extra sales before the other two react. But it doesn’t, because it knows that the long-run equilibrium would be equal shares at lower prices for all three firms.
By the logic of the cftc’s position, we should (perhaps must) infer that the firm’s failure to maximize its short-run profits demonstrates agreement with the other two firms; and of course an agreement is a cartel, illegal per se under the Sherman Act. Antitrust has the same formal structure as the commodities laws. An outcome produced by agreement is illegal; the identical outcome produced by the game-theoretic effects of repeat dealing is lawful. Do we then use the observed behavior— consistent with both agreement and interdependence — to infer agreement and forbid the conduct? The answer in antitrust law has been “no,” as we explained in JTC Petroleum. Perhaps the ofto has a good reason why the answer to the oligopolistic interdependence question in antitrust is wrong, or why it does not apply to commodities trading. My colleagues surmise that economic differences between futures markets and physicals markets justify a legal difference. 202 F.3d at 936. Perhaps so; perhaps not. Yet the Commission did not confront this question. Courts are limited to the justifications the agency itself gives. SEC v. Chenery Corp., 318 U.S. 80, 88, 63 S.Ct. 454, 87 L.Ed. 626, (1943). The cfto simply assumed that if given conduct would be unusual unless either (a) agreement had been reached, or (b) small numbers produce interdependence, then explanation (a) must be the right one even in a small-numbers case. It did not attempt to grapple with traders’ interdependence, the parallel to antitrust law, the holding of Stoller, or any related issue, and Chenery forbids us to fill those gaps.
Still, it is conceivable that the traders prearranged their freshening. Statistical analysis might show that the pattern of trades early in 1991 was abnormal (as the cfto asserted). Or consider the interdependence issue. One of the four traders *945began each round by buying the spread, greatly increasing his risk (unless the others cooperated by buying back), although each needed to sell the spread and then buy later. They say that they understood what had to happen, so there was no real risk. Well, that’s testable. Take a month in which the oftc does not suspect agreement and see whether some traders who end up freshening a position enter with buy orders. Or take such a month and determine what portion of freshening trades are symmetrical or confined to small groups. The cftc treated trades without profit as proof of agreement; but maybe all it shows is that in a world with a lot of trades, some days will look like this even though most don’t. (Similarly, if you flip a fair coin, it is even money that you’ll get 10 heads in a row before 1,000 flips are done.) If this pattern happens one closing month in 100, then all the oftc has proven here is that if you wait long enough you’ll find this pattern — but without finding any agreement. But if in months that the oftc deems normal no big trader ever leads by buying the spread, then the order in this case would be sustainable. A remand thus could be justified, but enforcement of this order cannot be.