Plaintiffs Mueller Comercial de Mexico, S. de R.L. de C.V., and Southland Pipe *1229Nipples Company, Inc.1 (collectively, “Mueller”) appeal from a decision of the Court of International Trade sustaining the United States Department of Commerce’s (“Commerce”) antidumping determination. We vacate and remand.
BACKGROUND
The Tariff Act of 1930 (the “Act”), as amended, permits Commerce to levy anti-dumping duties on goods “sold in the United States at less than ... fair value.” 19 U.S.C. § 1673. Antidumping duty orders are issued for imported merchandise that is sold in the United States below its fair value and materially injures or threatens to injure a domestic industry. Id. An anti-dumping duty reflects the amount by which the “normal value” of a product (typically, the home market price — the price of the merchandise when sold for consumption in the exporting country), 19 U.S.C. § 1677b(l), exceeds the “export price” of the merchandise. 19 U.S.C. §§ 1673, 1677(35)(A). This difference is called the dumping margin. The imposition of the antidumping duty, equal to the dumping margin, is intended to ensure that merchandise is not sold in the United States below its fair value.
Commerce periodically reviews and reassesses antidumping duties imposed in earlier proceedings. 19 U.S.C. § 1675(a). On November 2, 1992, Commerce published an antidumping duty order on certain circular welded non-alloy steel pipe from Mexico. On November 2, 2009, Commerce published a notice of opportunity to request an administrative review of the anti-dumping duty order. Commerce received requests for administrative review from appellant Mueller; Tuberia Nacional, S.A. de C.V. (“TUNA”); and Ternium Mexico, S.A. de C.V. (“Ternium”), and from defendant-appellees Allied Tube and Conduit Corporation and TMK IPSCO Tubulars.
On December 23, 2009, Commerce initiated an antidumping administrative review concerning the period spanning from November 1, 2008, to October 31, 2009, issuing questionnaires to three mandatory respondents: (1) Mueller, an exporter, which purchased the majority of its subject merchandise from TUNA and Ternium, (2) TUNA and (3) Ternium, both producers of subject merchandise. TUNA’s review was rescinded (because there were no direct shipments), and Ternium opted not to participate in its own margin calculation. As a result, Commerce drew an adverse inference against Ternium pursuant to 19 U.S.C. § 1677e(b), assigning an adverse facts available (“AFA”) dumping margin of 48.33 percent, “which is the highest calculated transaction-specific margin from the most recently-completed administrative review of this antidumping duty order in which a rate was calculated.” J.A. 63 (Preliminary Results). Ternium’s dumping margin is not at issue in this appeal.
For Commerce to calculate Mueller’s an-tidumping rate, it was required to determine the difference between the “normal value” of Mueller’s goods (typically “home market” price) and the “export price” at which Mueller’s goods were sold in the United States. 19 U.S.C. §§ 1677(35)(A), 1677b(a). The “normal value” is ordinarily the price at which the goods were first sold for consumption in the exporting country — in this case, in Mexico. Id. § 1677b(a)(l)(B)(i). Here, Mueller had sufficient volume of home market sales such that they could be used to calculate “normal value.” See id. § 1677b(a)(1)(B)(ii)(II). However, where an exporter’s home market price is less *1230than the cost of production for the goods it sells, Commerce “may” disregard the below cost sales to calculate “normal value.” Id. § 1677b(b)(l). Therefore, Commerce must determine the cost of production of the subject merchandise. Id. § 1677b(b)(3). Such production costs are normally “calculated based on the records of the exporter or producer of the merchandise, if such records are kept in accordance with the generally accepted accounting principles of the exporting country ... and reasonably reflect the costs associated with the production and sale of the merchandise.” Id. § 1677b(f)(1)(A). If the cost of production is greater than the home market price, home market sales below production cost may be disregarded in calculating the normal value.2 Id. §§ 1677b(b)(1)(A); 1677b(b)(2)(C).
Although Mueller fully cooperated with Commerce’s review, Mueller did not possess all of the production cost information necessary to calculate its antidumping margin. See 19 U.S.C. § 1677b(b)(3). To calculate cost of production, Commerce requested data directly from Mueller’s two principal suppliers, TUNA and Ternium.
TUNA fully cooperated with the data requests, reporting cost of production on a product-specific basis. However, Ternium did not “provide detailed product-specific calculations that allocate[d] costs based on product dimensions.” J.A. 47 (Memorandum from Mark Flessner, Case Analyst, Dep’t of Commerce, to Richard Weible, Office Director, Dep’t of Commerce, Certain Circular Welded Non-Alloy Steel Pipe from Mexico: Use of [AFA] for Final Results 2 (June 13, 2011)) (“AFA Mem.”) (internal quotation marks omitted). Terni-um stated that it did not provide the data because it was not “readily available.” As a result, Commerce did not have the data necessary to calculate margins that took into account cost differences associated with the different physical characteristics of the goods.
For its preliminary analysis, Commerce simply relied on the submitted data and calculated a weighted-average dumping margin of 4.81 percent for Mueller. But because Ternium did not submit necessary cost data before the time for a final determination, in making the final calculations, Commerce used “facts otherwise available” to calculate Mueller’s margin under 19 U.S.C. § 1677e(a) of the statute. Specifically, Commerce concluded that the production costs of the goods Mueller acquired from Ternium (data that was unavailable) were related to acquisition costs (data that was available). Commerce identified the three sales transactions between TUNA and Mueller made at the greatest discount to Mueller — where Mueller’s acquisition cost was the furthest below TUNA’s production cost. Commerce then inferred that all of Ternium’s pipe that was sold to Mueller involved this discount for acquisition cost. This enabled Commerce to calculate Ternium’s cost of production from Mueller’s cost of acquisition from Ternium. Although there were other sales transactions between TUNA and Mueller that were not discounted as significantly, Commerce chose not to use that data. In its Final Results, Commerce used data from the three transactions to calculate a new weighted-average dumping rate for Mueller of 19.81 percent.
On August 22, 2011, Mueller filed suit in the Court of International Trade (“Trade Court”) seeking to overturn Commerce’s Final Results, noting that Mueller had ful*1231ly cooperated and alleging that Commerce’s application of “Ternium’s AFA to its calculation of the margin for Mueller,” despite Mueller’s full cooperation with Commerce’s requests, was improper. J.A. 197. Mueller argued that, instead, Commerce should have calculated production costs using the entire TUNA data set.3 The Trade Court found that Commerce’s application of facts available was reasonable, and sustained the Final Results. Mueller appealed. We have jurisdiction pursuant to 28 U.S.C. § 1295(a)(5).
Disoussion
I
As discussed, Mueller’s antidumping rate was based on the difference between the “normal value” of the subject merchandise (typically, the “home market” price) and the “export price” of the goods sold in the United States. See 19 U.S.C. § 1677(35)(A). But Mueller’s data alone could not be used to determine whether its home market sales were below the cost of production for the goods. Commerce requested data from Mueller’s two primary suppliers, TUNA and Ternium, to calculate the cost of production for the subject merchandise. See 19 U.S.C. § 1677b(b)(3). Ternium did not provide product-specific cost data that would enable Commerce to calculate Ternium’s cost of production for each product, failing to account for different costs based on nominal pipe size and pipe wall thickness. Therefore, Commerce did not have sufficient information to calculate Mueller’s antidumping rate.
When Commerce is missing necessary data, the statute provides two options to secure data that can be used as a substitute for the missing information. See 19 U.S.C. § 1677e. The first is “facts otherwise available.” The statute provides:
(a) In general
If—
(1) necessary information is not available on the record, or
(2) an interested party or any other person—
(A) withholds information that has been requested by [Commerce] under this subtitle,
(B) fails to provide such information by the deadlines for submission of the information or in the form and manner requested ...
[Commerce] shall ... use the facts otherwise available in reaching the applicable determination under this subtitle.
19 U.S.C. § 1677e(a) (emphases added). The second is the “adverse facts available” approach. In this respect, the statute provides:
(b) Adverse Inferences
If [Commerce] finds that an interested party has failed to cooperate by not acting to the best of its ability to comply with a request for information from [Commerce], [Commerce], in reaching the applicable determination under this subtitle, may use an inference that is adverse to the interests of that party in selecting from the facts otherwise available. Such adverse inference may include reliance on information derived from—
(1) the petition,
(2) a final determination in the investigation under this subtitle,
(3) any previous review under section 1675 of this title or determination under section 1675b of this title, or
*1232(4) any other information placed on the record.
19 U.S.C. § 1677e(b) (emphasis added).
These two subsections have different purposes. Subsection 1677e(a) (“subsection (a)”) may be used whether or not any party has failed to cooperate fully with the agency in its inquiry. See Zhejiang DunAn Hetian Metal Co. v. United States, 652 F.3d 1333, 1346 (Fed.Cir.2011) (“ ‘[T]he mere failure of a respondent to furnish requested information — for any reason — requires Commerce to resort to other sources of information to complete the factual record....”’ (emphasis added) (quoting Nippon Steel v. United States, 337 F.3d 1373, 1381 (Fed.Cir.2003))). In contrast, subsection 1677e(b) (“subsection (b)”) authorizes an inference adverse to an interested party when “Commerce makes the separate determination that [the party] has failed to cooperate by not acting to the best of its ability.” Id. (quoting Nippon Steel, 337 F.3d at 1381) (internal quotation marks omitted). In this case, Mueller is a cooperating party, while Ternium is not.
II
Initially we note that there is no contention here that Commerce, acting primarily under subsection (a) in setting a margin for Mueller, erred in using TUNA’s data as a surrogate for Ternium’s missing data. Mueller’s primary complaint is that Commerce limited its analysis to a small and unfavorable subset of the TUNA data. As stated above, Commerce used the three highest-margin transactions instead of taking the total number of transactions from the TUNA cost of production data, which resulted in a higher normal value, and therefore, a higher dumping margin. Mueller argues Commerce arbitrarily cherry-picked the data to achieve this higher dumping margin.
We separately address the two rationales that Commerce used to justify its approach. Commerce relied on the two rationales in combination, not on either one as an independent ground. If one fails, as we conclude it does, Commerce’s ruling cannot stand. SEC v. Chenery Corp., 332 U.S. 194, 196-97, 67 S.Ct. 1575, 91 L.Ed. 1995 (1947).
First, Commerce concluded that the use of the adverse inference to calculate Terni-um’s surrogate production cost actually yielded the most accurate calculation of Mueller’s antidumping rate.4
Mueller argues that this rationale is arbitrary and capricious or not supported by substantial evidence. We agree.
There is no support for Commerce’s claim that using the three least-favorable TUNA transactions would produce the most accurate dumping margin for Mueller. Even calculating Mueller’s dumping margin based on the TUNA transactions where Mueller purchased the subject merchandise at below cost prices showed that Mueller received an average discount that *1233was approximately half of the Commerce rate. An analysis based on all of the TUNA data showed that Mueller received, as a weighted average, less than a ten percent discount on the subject merchandise. Finally, an unweighted average of all the TUNA data showed that overall, Mueller’s acquisition costs were higher than TUNA’s production costs. Commerce has not explained why using a larger data set would produce a less accurate dumping margin. Commerce’s rationale that Ternium would have cooperated if disclosing its actual costs to Commerce had been favorable to its interests does not support a conclusion that the particular TUNA data Commerce ultimately chose to rely on accurately estimated those costs. There is no showing that Ternium, in the hypothesized benefit calculus, could have anticipated that, if it chose non-disclosure of its actual costs, Commerce would rely on TUNA’s three least favorable transactions to calculate Mueller’s rate; indeed, there is no showing that Ternium would even have known what TUNA’s data contained, given Commerce’s obligation to keep TUNA’s data confidential. 19 C.F.R. §§ 351.105, 351.303-06. Therefore, we find that Commerce’s accuracy rationale for its calculation of Mueller’s antidumping rate was unsupported by substantial evidence.
Because Commerce’s calculation of Mueller’s rate relied in part on this accuracy rationale, this decision must be set aside. There is no contention that the use of the particular TUNA data relied on by Commerce was somehow required by the antidumping statute. See, e.g., ICC v. Bhd. of Locomotive Eng’rs, 482 U.S. 270, 283, 107 S.Ct. 2360, 96 L.Ed.2d 222 (1987); Koyo Seiko Co. v. United States, 95 F.3d 1094, 1101 (Fed.Cir.1996). However, a reversal is also not appropriate because, as we conclude below, Commerce’s second rationale provides a possible factor supporting the rate that Commerce adopted.
Commerce’s second rationale rested on policy considerations unrelated to accuracy of the determination to be made on the already-developed record. Commerce found that Mueller could and should have induced Ternium’s cooperation by refusing to do business with Ternium, and Ternium would not be sufficiently deterred if Mueller were unaffected by Ternium’s non-cooperation, stating that Ternium could otherwise evade its antidumping rate by funneling its goods through Mueller.5 We conclude that Commerce may rely on such policies as part of a margin determination for a cooperating party like Mueller, as long as the application of those policies is reasonable on the particular facts and the predominant interest in accuracy is properly taken into account as well.
This analysis is justified and required, even if Commerce is viewed as acting entirely under subsection (a) in determining Mueller’s rate. But Mueller argues that because these adverse inferences and related rationales are the same as those that support the use of AFA under subsection *1234(b), they cannot support a “facts otherwise available” determination under subsection (a). Mueller is mistaken. Subsection (a) does not provide for the specific facts that should be used as a gap-filling mechanism. The statute on its face does not preclude Commerce from relying on the same considerations under subsection (a) for an AFA determination as used under subsection (b). Under Chevron, Commerce’s interpretation of subsections (a) and (b) otherwise governs as long as it is reasonable and a permissible statutory construction. United States v. Eurodif S.A., 555 U.S. 305, 316, 129 S.Ct. 878, 172 L.Ed.2d 679 (2009); Timken Co. v. United States, 354 F.3d 1334, 1342 (Fed.Cir.2004). Consideration under subsection (a) of facts found or rationales applicable under subsection (b), in the way Commerce may be viewed as having done here, passes muster under Chevron.
This result is wholly consistent with our precedents applying subsection (b) itself, which is properly directed to non-cooperating parties. This Court’s decision in F.lli De Cecco Di Filippo Fara S. Martino S.p.A. v. United States, 216 F.3d 1027, 1032 (Fed.Cir.2000), required that, even for a non-cooperating party, subsection (b) be applied to arrive at “a reasonably accurate estimate of the respondent’s actual rate, albeit with some built-in increase intended as a deterrent to noncompliance.” All the more so for a cooperating party, for which the equities would suggest greater emphasis on accuracy in the overall mix. Moreover, this Court’s decision in Chang-zhou made clear that, in the case of a cooperating party, Commerce cannot confine itself to a deterrence rationale and also must carry out a case-specific analysis of the applicability of deterrence and similar policies. Changzhou Wujin Fine Chemical Factory Co., Ltd. v. U.S., 701 F.3d 1367, 1379 (Fed.Cir.2012). And those principles were in no way questioned in Fine Furniture (Shanghai) Ltd. v. United States, No. 2013-1158, 748 F.3d 1365, 1370-71, 2014 WL 1613883, at *4 (Fed.Cir. Apr. 23, 2014), which simply rejected a contention that a countervailing duty rate for a cooperating importer could not be based on adverse inferences drawn against a non-cooperating foreign country (about the country’s subsidizing of an input into the importer’s product). Fine Furniture involved no issue about the application of the De Ceceo and Changzhou analysis to the selection of the particular rate for the cooperating party.
Contrary to Mueller’s contention, consideration of various factors in calculating the rate of a cooperating party is not precluded by Changzhou. In Changzhou, Commerce concluded it was necessary to use an adverse rate against a cooperating party because other rates “would not be sufficiently adverse as to effectuate the purpose of the facts available rule to induce respondents to provide [Commerce] with complete and accurate information”— in other words, they would not have sufficient deterrent effect. Changzhou, 701 F.3d at 1378 (internal quotations omitted). But there was no support in the statute for imposing any deterrent effect on cooperating parties in that case. Id. at 1379. The cooperating parties could not have induced the non-cooperating party to provide complete and accurate information, thus “there was no need or justification for deterrence.” Id. Nor was there a claim that the non-cooperating party was likely to evade its own antidumping duty through the cooperating parties. Id. We concluded that it was unreasonable to rely on a deterrence rationale. We reversed and remanded to Commerce to “act non-arbitrarily” in calculating the separate rate for the cooperating parties. Id.
There is potentially greater support for Commerce’s use of an evasion or induce*1235ment rationale in this ease than in Chang-zhou. While the cooperating plaintiffs in Changzhou did not have any mechanism to force the non-cooperating party’s cooperation (since the cooperating parties did not purchase goods from the non-cooperating party), id. at 1370-71, Mueller had an existing relationship with supplier Ternium. Therefore, Mueller could potentially have refused to do business with Ternium in the future as a tactic to force Ternium to cooperate. In fact, the relationship between Mueller and Ternium is similar to the relationship between the importer and exporter in KYD, Inc. v. United States, 607 F.3d 760, 768 (Fed.Cir.2010). There, King Pac and KYD had an existing relationship as importer-exporter, and this court found that KYD could have used this relationship to induce King Pac to cooperate. Id. (“In the aggregate, however, the importers’ exposure to enhanced anti-dumping duties seems likely to have the effect of either directly inducing cooperation from the exporters with whom the importers deal or doing so indirectly, by leaving uncooperative exporters without importing partners who are willing to deal in their products.”); see also Fine Furniture, 2014 WL 1613883, at *7 (“Fine Furniture is a company within the country of China, benefitting directly from the subsidies the government of China may be providing [and] a remedy that collaterally reaches Fine Furniture has the potential to encourage the government of China to cooperate so as to not hurt its overall industry.”). So too with Mueller and Ter-nium — if Mueller and other entities were not willing to export goods produced by Ternium, this would potentially induce Ternium to cooperate. On the other hand, if the cooperating entity has no control over the non-cooperating suppliers, a resulting adverse inference is potentially unfair to the cooperating party. SKF USA Inc. v. United States, 630 F.3d 1365, 1375 (Fed.Cir.2011).
In addition, as Commerce recognized, there is the possibility that Ternium could evade its own AFA rate of 48.33 percent by exporting its goods through Mueller if Mueller were assigned a favorable dumping rate. In this respect, too, this case is different from Changzhou and similar to KYD. We noted there that “KYD’s argument would allow an uncooperative foreign exporter to avoid the adverse inferences permitted by statute simply by selecting an unrelated importer, resulting in easy evasion of the means Congress intended for Commerce to use to induce cooperation with its antidumping investigations.” KYD, 607 F.3d at 768. Mueller argued that there was no evidence that Mueller was likely to act on Ternium’s behalf and reasoned that Commerce could investigate such false exports and impose Ternium’s own antidumping rate on them. See, e.g., Tung Mung Dev. Co. v. United States, 354 F.3d 1371, 1381 n. 10 (Fed.Cir.2004). But the fact that Commerce has alternative methods for addressing evasion does not mean that the particular chosen method is arbitrary. Commerce can use all of the methods provided in the Act for enforcement of the antidumping provisions.
Ill
In summary, on the remand, Commerce should recalculate Mueller’s rate. In doing so, Commerce must have as its primary objective the calculation of an accurate rate for Mueller — as we said in Changzhou — “[w]e find no support in- our caselaw or the statute’s plain text for the proposition that deterrence, rather than fairness or accuracy, is the overriding purpose of the antidumping statute when calculating a rate for a cooperating party.” 701 F.3d at 1378 (internal quotation marks omitted). But we do not foreclose *1236Commerce from also relying on the policy considerations that motivated the decision under review — namely, its desire to encourage Mueller to induce Ternium’s cooperation and Commerce’s concern that calculating too low a rate for Mueller might allow Ternium to evade its own dumping duty by channeling sales through Mueller.
Commerce must take into account that Mueller itself was a cooperating party and that Commerce’s inducement/evasion approach to Mueller’s rate calculation could discourage Mueller’s own cooperation. See id. To the extent that Commerce chooses to rely on inducement/evasion considerations, its approach must be reasonable. We do not today decide whether relying on inducement/evasion rationales to calculate Mueller’s rate would be reasonable in the circumstances of this case. We only hold that the statute does not preclude reliance on inducement or evasion considerations in calculating Mueller’s rate, and that such an approach is not foreclosed by Changzhou. We leave it to Commerce in the first instance to determine the relevant considerations and balance the need to calculate an accurate rate for Mueller and Mueller’s status as a cooperating party with other potentially relevant concerns.
Finally, we wish to be clear that under subsection (b) we do not bar Commerce from drawing adverse inferences against a non-cooperating party that have collateral consequences for a cooperating party. Where an adverse inference is used to calculate the rate of a non-cooperating party that rate may sometimes be used in calculating the rate of a cooperating party and thus have collateral consequences for the cooperating party. KYD, 607 F.3d at 768. That is not the situation here. Commerce drew two adverse inferences against Ternium. The first adverse inference was used to calculate Ternium’s own antidump-ing rate of 48.33 percent under subsection (b). The second adverse inference against Ternium, used to approximate Ternium’s cost of production, was not used in calculating Ternium’s rate, but only in calculating Mueller’s rate. So too this is unlike Fine Furniture where the government of China provided a subsidy to Fine Furniture. Fine Furniture, 748 F.3d at 1371-73, 2014 WL 1613883, at *5-6. China was an “interested party” as defined by the statute and the adverse inference applied was “adverse to the interests of that party.” 19 U.S.C. § 1677(9)(B); see also Fine Furniture, 748 F.3d at 1370-71, 2014 WL 1613883, at *4. The use of an adverse inference was contrary to the interest of China because it directly offset the subsidy that China provided. Here, there is no direct adverse effect on Ternium from using an adverse inference as facts otherwise available in computing Mueller’s dumping margin. Under these circumstances Commerce must proceed in the manner we have described.
VACATED AND REMANDED.
Costs
No costs.