NACS v. Board of Governors of the Federal Reserve System

 United States Court of Appeals
          FOR THE DISTRICT OF COLUMBIA CIRCUIT



Argued January 17, 2014                Decided March 21, 2014

                         No. 13-5270

  NACS, FORMERLY KNOWN AS NATIONAL ASSOCIATION OF
            CONVENIENCE STORES, ET AL.,
                    APPELLEES

                               v.

 BOARD OF GOVERNORS OF THE FEDERAL RESERVE SYSTEM,
                    APPELLANT


         Appeal from the United States District Court
                 for the District of Columbia
                     (No. 1:11-cv-02075)


     Katherine H. Wheatley, Associate General Counsel, Board
of Governors of the Federal Reserve System, argued the cause
for appellant. With her on the briefs were Richard M. Ashton,
Deputy General Counsel, Yvonne F. Mizusawa, Senior Counsel,
and Joshua P. Chadwick, Counsel.

     Seth P. Waxman argued the cause for amici curiae The
Clearing House Association, L.L.C., et al. in support of neither
party. With him on the brief were Albinas Prizgintas, Noah A.
Levine, and Alan Schoenfeld.
                               2
     Shannen W. Coffin argued the cause for appellees. With him
on the brief was Linda C. Bailey.

    Andrew G. Celli Jr., Ilann M. Maazel, and O. Andrew F.
Wilson were on the brief for amicus curiae The Retail Litigation
Center, Inc. in support of appellees.

    Jeffrey I. Shinder was on the brief for amici curiae
7-Eleven, Inc., et al. in support of appellees.

     David A. Balto was on the brief for amicus curiae United
States Senator Richard J. Durbin in support of appellees.

   Before: TATEL, Circuit Judge, and EDWARDS and
WILLIAMS, Senior Circuit Judges.

    Opinion for the Court filed by Circuit Judge TATEL.

     TATEL, Circuit Judge: Combining features of credit cards
and checks, debit cards have become not just the most popular
noncash payment method in the United States but also a source
of substantial revenue for banks and companies like Visa and
MasterCard that own and operate debit card networks. In 2009
alone, debit card holders used their cards 37.6 billion times,
completing transactions worth over $1.4 trillion and yielding
over $20 billion in fees for banks and networks. Concerned that
these fees were excessive and that merchants, who pay the fees
directly, and consumers, who pay a portion of the fees indirectly
in the form of higher prices, lacked any ability to resist them,
Congress included a provision in the Dodd-Frank financial
reform act directing the Board of Governors of the Federal
Reserve System to address this perceived market failure. In
response, the Board issued regulations imposing a cap on the
per-transaction fees banks receive and, in an effort to force
                                3
networks to compete for merchants’ business, requiring that at
least two networks owned and operated by different companies
be able to process transactions on each debit card. Merchant
groups challenged the regulations, seeking lower fees and even
more network competition. The district court granted summary
judgment to the merchants, concluding that the rules violate the
statute’s plain language. We disagree. Applying traditional tools
of statutory interpretation, we hold that the Board’s rules
generally rest on reasonable constructions of the statute, though
we remand one minor issue—the Board’s treatment of so-called
transactions-monitoring costs—to the Board for further
explanation.

                                I.
     Understanding this case requires looking under the hood—
or, more accurately, behind the teller’s window—to see what
really happens when customers use their debit cards. After
providing some background about debit cards and the debit card
marketplace, we outline Congress’s effort to solve several
perceived market failures, the Board’s attempt to put Congress’s
directives into action, and the district court’s rejection of the
Board’s approach.

                               A.
     We start with the basics. For purposes of this case, the term
“debit card” describes both traditional debit cards, which allow
cardholders to deduct money directly from their bank accounts,
and prepaid cards, which come loaded with a certain amount of
money that cardholders can spend down and, in some cases,
replenish. Debit card transactions are typically processed using
what is often called a “four party system.” The four parties are
the cardholder who makes the purchase, the merchant who
accepts the debit card payment, the cardholder’s bank (called the
“issuer” because it issues the debit card to the cardholder), and
                                4
the merchant’s bank (called the “acquirer” because it acquires
funds from the cardholder and deposits those funds in the
merchant’s account). In addition, each debit transaction is
processed on a particular debit card “network,” often affiliated
with MasterCard or Visa. The network transmits information
between the cardholder/issuer side of the transaction and the
merchant/acquirer side. Issuers activate certain networks on
debit cards, and only activated networks can process transactions
on those cards.

     Virtually all debit card transactions fall into one of two
categories: personal identification number (PIN) or signature.
PIN and signature transactions employ different methods of
“authentication”—a process that establishes that the cardholder,
and not a thief, has actually initiated the transaction. In PIN
authentication, the cardholder usually enters her PIN into a
terminal. In signature authentication, the cardholder usually
signs a copy of the receipt. Most networks can process either
PIN transactions or signature transactions, but not both.
Signature networks employ infrastructure used to process credit
card payments, while PIN networks employ infrastructure used
by ATMs. Only about one-quarter of merchants currently accept
PIN debit. Some merchants have never acquired the terminals
needed for customers to enter their PINs, while others believe
that signature debit better suits their business needs. More about
this later. And merchants who sell online generally refuse to
accept PIN debit because customers worry about providing PINs
over the Internet. Merchants who do accept both PIN and
signature debit often allow customers to select whether to
process particular transactions on a PIN network or a signature
network.

    Whether PIN or signature, a debit card transaction is
processed in three stages: authorization, clearance, and
                                5
settlement. Authorization begins when the cardholder swipes her
debit card, which sends an electronic “authorization request” to
the acquirer conveying the cardholder’s account information and
the transaction’s value. The acquirer then forwards that request
along the network to the issuer. Once the issuer has determined
whether the cardholder has sufficient funds in her account to
complete the transaction and whether the transaction appears
fraudulent, it sends a response to the merchant along the network
approving or rejecting the transaction. Even if the issuer
approves the transaction, that transaction still must be cleared
and settled before any money changes hands.

     Clearance constitutes a formal request for payment sent
from the merchant on the network to the issuer. PIN transactions
are authorized and cleared simultaneously: because the
cardholder generally enters her PIN immediately after swiping
her card, the authorization request doubles as the clearance
message. Signature transactions are first authorized and
subsequently cleared: because the cardholder generally signs
only after the issuer has approved the transaction, the merchant
must send a separate clearance message. This difference between
PIN and signature processing explains why certain businesses,
including car rental companies, hotels, and sit-down restaurants,
often refuse to accept PIN debit. Car rental companies authorize
transactions at pick-up to ensure that customers have enough
money in their accounts to pay but postpone clearance to allow
for the possibility that the customer might damage the vehicle or
return it without a full tank of gas. Hotels authorize transactions
at check-in but postpone clearance to allow for the possibility
that the guest might trash the room, order room service, or
abscond with the towels and robes. And sit-down restaurants
authorize transactions for the full amount of the meal but
postpone clearance to give diners an opportunity to add a tip.
                                6
     The final debit card payment processing step, settlement,
involves the actual transfer of funds from the issuer to the
acquirer. After settlement, the cardholder’s account has been
debited, the merchant’s account has been credited, and the
transaction has concluded. Rather than settle transactions one-
by-one, banks generally employ companies that determine each
bank’s net debtor/creditor position over a large number of
transactions and then settle those transactions simultaneously.

     Along the way, and central to this case, the parties charge
each other various fees. The issuer charges the acquirer an
“interchange fee,” sometimes called a “swipe fee,” which
compensates the issuer for its role in processing the transaction.
The network charges both the issuer and the acquirer “network
processing fees,” otherwise known as “switch fees,” which
compensate the network for its role in processing the
transaction. Finally, the acquirer charges the merchant a
“merchant discount,” the difference between the transaction’s
face value and the amount the acquirer actually credits the
merchant’s account. Because the merchant discount includes the
full value of the interchange fee, the acquirer’s portion of the
network processing fee, other acquirer and network costs, and a
markup, merchants end up paying most of the costs acquirers
and issuers incur. Merchants in turn pass some of these costs
along to consumers in the form of higher prices. In contrast to
credit card fees, which generally represent a set percentage of
the value of a transaction, debit card fees change little as price
increases. Thus, a bookstore might pay the same fees to sell a
$25 hardcover that Mercedes would pay to sell a $75,000 car.

     Before the Board promulgated the rules challenged in this
case, networks and issuers took advantage of three quirks in the
debit card market to increase fees without losing much business.
First, issuers had complete discretion to decide whether to
                                7
activate certain networks on their cards. For instance, an issuer
could limit payment processing to one Visa signature network, a
Visa signature network and a Visa PIN network, or Visa and
MasterCard signature and PIN networks. Second, networks had
complete discretion to set the level of interchange and network
processing fees. Finally, Visa and MasterCard controlled most of
the debit card market. According to one study entered into the
record, in 2009 networks affiliated with Visa or MasterCard
processed over eighty percent of all debit transactions. Steven C.
Salop, et al., Economic Analysis of Debit Card Regulation
Under Section 920, Paper for the Board of Governors of the
Federal Reserve System 10 (Oct. 27, 2010). Making things
worse for merchants, these companies imposed “Honor All
Cards” rules that prohibited merchants from accepting some but
not all of their credit cards and signature debit cards. Merchants
were therefore stuck paying whatever fees Visa and MasterCard
chose to set, unless they refused to accept any Visa and
MasterCard credit and signature debit cards—hardly a realistic
option for most merchants given the popularity of plastic.

     Exercising this market power, issuers and networks often
entered into mutually beneficial agreements under which issuers
required merchants to route transactions on certain networks that
generally charged high processing fees so long as those networks
also set high interchange fees. Many of these agreements were
exclusive, meaning that issuers agreed to activate only one
network or only networks affiliated with one company.
Networks and issuers also negotiated routing priority
agreements, which forced merchants to process transactions on
certain activated networks rather than others. By 2009,
interchange and network processing fees had reached, on
average, 55.5 cents per transaction, including a 44 cent
interchange fee, a 6.5 cent network processing fee charged to the
issuer, and a 5 cent network processing fee charged to the
                                8
acquirer. Debit Card Interchange Fees and Routing, Notice of
Proposed Rulemaking (“NPRM”), 75 Fed. Reg. 81,722, 81,725
(Dec. 28, 2010).

                                B.
     Seeking to correct the market defects that were contributing
to high and escalating fees, Congress passed the Durbin
Amendment as part of the 2010 Dodd-Frank Wall Street Reform
and Consumer Protection Act, Pub. L. No. 111-203, 124 Stat.
1376 (2010). The amendment, which modified the Electronic
Funds Transfer Act (EFTA), Pub. L. No. 95-630, 92 Stat. 3641
(1978), contains two key provisions. The first, EFTA section
920(a), restricts the amount of the interchange fee. Specifically,
it instructs the Board of Governors of the Federal Reserve
System to promulgate regulations ensuring that “the amount of
any interchange transaction fee . . . is reasonable and
proportional to the cost incurred by the issuer with respect to the
transaction.” 15 U.S.C. § 1693o-2(a)(3)(A); see also id. §
1693o-2(a)(6)–(7)(A) (exempting debit cards issued by banks
that, combined with all affiliates, have assets of less than $10
billion and debit cards affiliated with certain government
payment programs from interchange fee regulations). To this
end, section 920(a)(4)(B), in language the parties hotly debate,
requires the Board to “distinguish between . . . the incremental
cost incurred by an issuer for the role of the issuer in the
authorization, clearance, or settlement of a particular debit
transaction, which cost shall be considered . . . , [and] other
costs incurred by an issuer which are not specific to a particular
electronic debit transaction, which costs shall not be
considered.” Id. § 1693o-2(a)(4)(B)(i)-(ii). Like the parties, we
shall refer to the costs of “authorization, clearance, and
settlement” as “ACS costs.” In addition, section 920(a) “allow[s]
for an adjustment to the fee amount received or charged by an
issuer” to compensate for “costs incurred by the issuer in
                                9
preventing fraud in relation to electronic debit transactions
involving that issuer,” so long as the issuer “complies with the
fraud-related standards established by the Board.” Id. § 1693o-
2(a)(5)(A).

      The second key provision, EFTA section 920(b), prohibits
certain exclusivity and routing priority agreements. Specifically,
it instructs the Board to promulgate regulations preventing any
“issuer or payment card network” from “restrict[ing] the number
of payment card networks on which an electronic debit
transaction may be processed to . . . 1 such network; or . . . 2 or
more [affiliated networks].” Id. § 1693o-2(b)(1)(A). It also
directs the Board to prescribe regulations that prohibit issuers
and networks from “inhibit[ing] the ability of any person who
accepts debit cards for payments to direct the routing of
electronic debit transactions for processing over any payment
card network that may process such transactions.” Id. § 1693o-
2(b)(1)(B). Congress anticipated that these prohibitions would
force networks to compete for merchants’ business, thus driving
down fees.

                                C.
     In late 2010, the Board proposed rules to implement
sections 920(a) and (b). As for section 920(a), the Board
proposed allowing issuers to recover only “incremental” ACS
costs and interpreted “incremental” ACS costs to mean costs
that “vary with the number of transactions” an issuer processes
over the course of a year. NPRM, 75 Fed. Reg. at 81,735. Issuers
would thus be unable to recover “costs that are common to all
debit card transactions and could never be attributed to any
particular transaction (i.e., fixed costs), even if those costs are
specific to debit card transactions as a whole.” Id. at 81,736. The
Board “recognize[d]” that this definition would “impose[] a
burden on issuers by requiring issuers to segregate costs that
                                10
vary with the number of transactions from those that are largely
invariant to the number of transactions” and “that excluding
fixed costs may prevent issuers from recovering through
interchange fees some costs associated with debit card
transactions.” Id. The Board nonetheless determined that other
definitions of “incremental cost” “do not appropriately reflect
the incremental cost of a particular transaction to which the
statute refers.” Id. at 81,735. Limiting the interchange fee to
average variable ACS costs, the Board proposed allowing
issuers to recover at most 12 cents per transaction—considerably
less than the 44 cents issuers had previously received on
average. Id. at 81,736–39.

     After evaluating thousands of comments, the Board issued a
Final Rule that almost doubled the proposed cap. The Board
abandoned its proposal to define “incremental” ACS costs to
mean average variable ACS costs, deciding instead not to define
the term “incremental costs” at all. Debit Card Interchange Fees
and Routing, Final Rule (“Final Rule”), 76 Fed. Reg. 43,394,
43,426–27 (July 20, 2011). Observing that “the requirement that
one set of costs be considered and another set of costs be
excluded suggests that Congress left to the implementing agency
discretion to consider costs that fall into neither category to the
extent necessary and appropriate to fulfill the purposes of the
statute,” the Board allowed issuers to recover all costs “other
than prohibited costs.” Id. Thus, in addition to average variable
ACS costs, issuers could recover: (1) what the proposed rule had
referred to as “fixed” ACS costs; (2) costs issuers incur as a
result of transactions-monitoring to prevent fraud; (3) fraud
losses, which are costs issuers incur as a result of settling
fraudulent transactions; and (4) network processing fees. Id. at
43,429–31. The Board prohibited issuers from recovering other
costs, such as corporate overhead and debit card production and
delivery costs, that the Board determined were not incurred to
                                11
process specific transactions. Id. at 43,427–29. Accounting for
all permissible costs, the Board raised the interchange fee cap to
21 cents plus an ad valorem component of 5 basis points (.05
percent of a transaction’s value) to compensate issuers for fraud
losses. Id. at 43,404.

     In response to section 920(b), the Board’s proposed rule
outlined two possible approaches. Under “Alternative A,”
issuers would have to activate at least two unaffiliated networks
on each debit card regardless of method of authentication.
NPRM, 75 Fed. Reg. at 81,749. For example, an issuer could
activate a Visa signature network and a MasterCard PIN
network. Under “Alternative B,” issuers would have to activate
at least two unaffiliated networks for each method of
authentication. Id. at 81,749–50. For example, an issuer could
activate both Visa and MasterCard signature and PIN networks.

     In the Final Rule the Board chose Alternative A.
Acknowledging that “Alternative A provides merchants fewer
routing options,” the Board reasoned that it satisfied statutory
requirements and advanced Congress’s desire to enhance
competition among networks without excessively undermining
the ability of cardholders to route transactions on their preferred
networks or “potentially limit[ing] the development and
introduction of new authentication methods.” Final Rule, 76
Fed. Reg. at 43,448.

                                D.
     Upset that the Board had nearly doubled the interchange fee
cap (as compared to the proposed rule) and had selected the less
restrictive anti-exclusivity option, several merchant groups,
including NACS, the organization formerly known as the
National Association of Convenience Stores, filed suit in district
court. The merchants argued that both rules violate the plain
                               12
terms of the Durbin Amendment: the interchange fee cap
because the statute allows issuers to recover only average
variable ACS costs, not “fixed” ACS costs, transactions-
monitoring costs, fraud losses, or network processing fees; and
the anti-exclusivity rule because the statute requires that all
merchants—even those who refuse to accept PIN debit—be able
to route each debit transaction on multiple unaffiliated networks.
Several financial services industry groups, which during
rulemaking had urged the Board to set an even higher
interchange fee cap and adopt an even less restrictive anti-
exclusivity rule, participated as amici curiae in support of
neither party.

     The district court granted summary judgment to the
merchants. The court began by observing that “[a]ccording to
the Board, [the statute contains] ambiguity that the Board has
discretion to resolve. How convenient.” NACS v. Board of
Governors of the Federal Reserve System, 958 F. Supp. 2d 85,
101 (D.D.C. 2013). Rejecting this view, the district court
determined that the Durbin Amendment is “clear with regard to
what costs the Board may consider in setting the interchange fee
standard: Incremental ACS costs of individual transactions
incurred by issuers may be considered. That’s it!” Id. at 105. The
district court thus concluded that the Board had erred in
allowing issuers to recover “fixed” ACS costs, transactions-
monitoring costs, fraud losses, and network processing fees. Id.
at 105–09. The court also agreed with the merchants that section
920(b) unambiguously requires that all merchants be able to
route every transaction on at least two unaffiliated networks. Id.
at 109–14. The Board’s final anti-exclusivity rule, the district
court held, “not only fails to carry out Congress’s intention; it
effectively countermands it!” Id. at 112. Concluding that “the
Board completely misunderstood the Durbin Amendment’s
statutory directive and interpreted the law in ways that were
                                13
clearly foreclosed by Congress,” the district court vacated and
remanded both the interchange fee rule and the anti-exclusivity
rule. Id. at 114. But because regulated parties had already “made
extensive commitments” in reliance on the Board’s rules, the
district court stayed vacatur to provide the Board a short period
of time in which to promulgate new rules consistent with the
statute. Id. at 115. Subsequently, the district court granted a stay
pending appeal.

     The Board now appeals, arguing that both rules rest on
reasonable constructions of ambiguous statutory language.
Financial services amici, urging reversal but still ostensibly
appearing in support of neither party, filed a brief and
participated in oral argument—though we have considered only
those arguments that at least one party has not disavowed. See
Eldred v. Reno, 239 F.3d 372, 378 (D.C. Cir. 2001) (noting that
arguments “rejected by the actual parties to this case” are “not
properly before us”); Eldred v. Ashcroft, 255 F.3d 849, 854
(D.C. Cir. 2001) (Sentelle, J., dissenting from denial of
rehearing en banc) (“Under the panel’s holding, it is now the law
of this circuit that amici are precluded both from raising new
issues and from raising new arguments.”). In a case like this, “in
which the District Court reviewed an agency action under the
[Administrative Procedures Act], we review the administrative
action directly, according no particular deference to the
judgment of the District Court.” In re Polar Bear Endangered
Species Act Listing and Section 4(d) Rule Litigation, 720 F.3d
354, 358 (D.C. Cir. 2013) (internal quotation marks omitted).
Because the Board has sole discretion to administer the Durbin
Amendment, we apply the familiar two-step framework set forth
in Chevron U.S.A. Inc. v. Natural Resources Defense Council,
Inc., 467 U.S. 837 (1984). At Chevron’s first step, we consider
whether, as the district court concluded, Congress has “directly
spoken to the precise question at issue.” Id. at 842. If not, we
                                14
proceed to Chevron’s second step where we determine whether
the Board’s rules rest on “reasonable” interpretations of the
Durbin Amendment. Id. at 844.

     Before addressing the parties’ arguments, we think it worth
emphasizing that Congress put the Board, the district court, and
us in a real bind. Perhaps unsurprising given that the Durbin
Amendment was crafted in conference committee at the eleventh
hour, its language is confusing and its structure convoluted. But
because neither agencies nor courts have authority to disregard
the demands of even poorly drafted legislation, we must do our
best to discern Congress’s intent and to determine whether the
Board’s regulations are faithful to it.

                                II.
     We begin with the interchange fee. Recall that section
920(a)(4)(B)(i) requires the Board to include “incremental
cost[s] incurred by an issuer for the role of the issuer in the
authorization, clearance, or settlement of a particular electronic
debit transaction,” and that section 920(a)(4)(B)(ii) prohibits the
Board from including “other costs incurred by an issuer which
are not specific to a particular electronic debit transaction.”
Echoing the district court, the merchants argue that the two
sections unambiguously permit issuers to recover only
“incremental” ACS costs. “The plain language of the Durbin
Amendment,” the merchants insist, “does not grant the Board
the discretion it claims to consider costs beyond those delineated
in Section 920(a)(4)(B).” Appellees’ Br. 26; see also NACS, 958
F. Supp. 2d at 100 (noting that the district court had “no
difficulty concluding that the statutory language evidences an
intent by Congress to bifurcate the entire universe of costs
associated with interchange fees”). Alternatively, the merchants
briefly argue that even if section 920(a)(4)(B) is ambiguous, the
Board’s resolution of that ambiguity was unreasonable—though
                                15
they acknowledge that this argument essentially rehashes their
Chevron step one argument. See Appellees’ Br. 44 (“Many of
the same arguments discussed above also demonstrate the
unreasonableness of the interchange fee standard.”). The Board
also thinks the Durbin Amendment is unambiguous, though it
argues that the statute clearly establishes a third category of
costs: those that are not “incremental” ACS costs but are
specific to a particular transaction. See Final Rule, 76 Fed. Reg.
at 43,426 (“[T]here exist costs that are not encompassed in
either the set of costs the Board must consider under Section
920(a)(4)(B)(i), or the set of costs the Board may not consider
under Section 920(a)(4)(B)(ii).”). Relying on the requirement
that the interchange transaction fee be “reasonable and
proportional to the cost incurred by the issuer with respect to the
transaction,” 15 U.S.C. § 1693o-2(a)(2), (a)(3)(A), the Board
concludes that it may but need not allow issuers to recover costs
falling within this third category, subject of course to other
statutory constraints. Like the merchants, the Board also offers a
Chevron step two argument. See Appellant’s Br. 71 (“Even
assuming for the sake of argument that the district court offered
a possible reading, the statute does not unambiguously foreclose
the Board’s construction . . . .”).

    The parties’ competing arguments present us with two
options. Were we to agree with the merchants that the statute
allows recovery only of “incremental” ACS costs, we would
have to invalidate the rule without considering the particular
categories of costs the merchants challenge given that the Board
expressly declined to define the ambiguous statutory term
“incremental,” let alone determine whether those particular types
of costs qualify as “incremental” ACS costs. See Securities &
Exchange Commission v. Chenery Corp., 318 U.S. 80, 87 (1943)
(“The grounds upon which an administrative order must be
judged are those upon which the record discloses that its action
                                 16
was based.”). Were we to determine that the Board’s reading of
section 920(a)(4)(B) is either compelled by the statute or
reasonable, we would have to go on to consider whether the
statute allows recovery of “fixed” ACS costs, transactions-
monitoring costs, fraud losses, and network processing fees. We
must therefore first decide whether section 920(a)(4)(B)
bifurcates the entire universe of costs the Board may consider, or
whether the statute allows for the existence of a third category of
costs that falls outside the two categories specifically listed.

                                 A.
     The Board may well have been able to interpret section
920(a)(4)(B) as the merchants urge. Such a reading could rely on
the statutory mandate to “distinguish between” one set of costs
and “other costs,” and could interpret section 920(a)(4)(B)(i) as
referring to variable costs and section 920(a)(4)(B)(ii) as
referring to fixed costs. But contrary to the merchants’ position,
and consistent with the Board’s Chevron step two argument, we
certainly see nothing in the statute’s language compelling that
result. The merchants’ preferred reading requires assuming that
the phrase “incremental cost incurred by the issuer for the role of
the issuer in the authorization, clearance, and settlement of a
particular electronic debit transaction” describes all issuer costs
“specific to a particular electronic debit transaction.” For several
reasons, however, we believe that phrase could just as easily, if
not more easily, be read to qualify the language of section
920(a)(4)(B)(i) such that it encompasses a subset of costs
specific to a particular transaction, leaving other costs specific to
a particular transaction unmentioned.

     To begin with, as the Board pointed out in the Final Rule,
the phrase “incremental cost” has a several possible definitions,
including marginal cost, variable cost, “the cost of producing
some increment of output greater than a single unit but less than
                                 17
the entire production run,” and “the difference between the cost
incurred by a firm if it produces a particular quantity of a good
and the cost incurred by the firm if it does not produce the good
at all.” Final Rule, 76 Fed. Reg. at 43,426–27. As a result,
depending on how these terms are defined, the category of
“incremental” costs would not necessarily encompass all costs
that are “specific to a particular electronic debit transaction.” See
infra at 26 (noting the parties’ agreement that the “specific to a
particular electronic debit transaction” phrase should not be read
to limit issuers to recovering only the marginal cost of each
particular transaction).

     Second, the phrase “incurred by an issuer for the role of the
issuer in the authorization, clearance, or settlement of a
particular electronic debit transaction” limits the class of
“incremental” costs the Board must consider. So even if the
word “incremental” were read to include all costs specific to a
particular transaction, Congress left unmentioned incremental
costs other than incremental ACS costs. See Final Rule, 76 Fed.
Reg. at 43,426 n.116 (“The reference in Section 920(a)(4)(B)(i)
requiring consideration of the incremental costs incurred in the
‘authorization, clearance, or settlement of a particular
transaction’ and the reference in section 920(a)(4)(B)(ii)
prohibiting consideration of costs that are ‘not specific to a
particular electronic debit transaction,’ read together, recognize
that there may be costs that are specific to a particular electronic
debit transaction that are not incurred in the authorization,
clearance, or settlement of that transaction.”). For example, in
the proposed rule the Board determined that “cardholder rewards
that are paid by the issuer to the cardholder for each transaction”
and “costs associated with providing customer service to
cardholders for particular transactions” are “associated with a
particular transaction” but “are not incurred by the issuer for its
role in authorization, clearing, and settlement of that
                                 18
transaction.” NPRM, 75 Fed. Reg. at 81,735. Moreover, in the
Final Rule the Board explained that fraud losses “are specific to
a particular transaction” because they result from the settlement
of particular fraudulent transactions, but are not incurred by the
issuer for the role of the issuer in the authorization, clearance, or
settlement of particular transactions. Final Rule, 76 Fed. Reg. at
43,431 (describing fraud losses as “the result of an issuer’s
authorization, clearance, or settlement of a particular electronic
debit transaction that the cardholder later identifies as
fraudulent”); see also Appellant’s Br. 67 (defending the Board’s
decision to allow issuers to recover some fraud losses on the
ground that fraud losses fall outside section 920(a)(4)(B)).

     Third, as the Board pointed out, had Congress wanted to
allow issuers to recover only incremental ACS costs, it could
have done so directly. See Final Rule, 76 Fed. Reg. at 43,426.
For instance, in section 920(a)(3)(A) Congress could have
instructed the Board to “promulgate regulations ensuring that
interchange fees are reasonable and proportional to the
incremental costs of authorization, clearance, and settlement that
an issuer incurs with respect to a particular electronic debit
transaction.” Instead, in section 920(a)(3)(A) Congress required
the Board to promulgate regulations ensuring that interchange
fees are “reasonable and proportional to the cost incurred by the
issuer with respect to the transaction” and separately instructed
the Board, when determining issuer costs, to “distinguish
between” incremental ACS costs, which the Board must
consider, 15 U.S.C. § 1693o-2(a)(4)(B)(i), and “other costs . . .
which are not specific to a particular electronic debit
transaction,” which the Board must not consider, id. § 1693o-
2(a)(4)(B)(ii).

    The merchants advance several arguments in support of the
opposite conclusion. They first assert that the “which” clause in
                                19
the phrase “other costs incurred by an issuer which are not
specific to a particular electronic debit transaction” should be
read descriptively rather than restrictively. As their labels
suggest, descriptive clauses explain, while restrictive clauses
define. To illustrate, consider a simple sentence: “the cars which
are blue have sunroofs.” Read descriptively, the clause “which
are blue” states a fact about the entire class of cars, which also
happen to have sunroofs. Read restrictively, the clause defines a
particular class of cars—blue cars—all of which have sunroofs.
Although often subtle, the distinction between descriptive and
restrictive clauses makes all the difference in this case. Here’s
why.

     We have thus far assumed that section 920(a)(4)(B)(ii)’s
“which” clause should be read restrictively. On this reading (the
Board’s), the clause defines the class of “other costs” issuers are
precluded from recovering. As explained above, based on this
restrictive reading the Board reasonably concluded that the
statute establishes three categories of costs. But if the clause
should instead be read descriptively, then it would describe a
characteristic of “other costs” without limiting the meaning of
“other costs.” On this reading (the merchants’), the statute
bifurcates the entire universe of costs, requiring the Board to
define the statutory term “incremental cost incurred by an issuer
for the role of the issuer in the authorization, clearance, or
settlement of a particular electronic debit transaction” as
including all costs other than costs “not specific to a particular
electronic debit transaction.”

     Normally, writers distinguish between descriptive and
restrictive clauses by setting the former but not the latter aside
with commas and by introducing the former with “which” and
the latter with “that.” Here, Congress introduced the clause at
issue with the word “which” but failed to set it aside with
                                 20
commas. Word choice thus suggests a descriptive reading of the
clause, while punctuation suggests a restrictive reading. In
support of a descriptive reading, the merchants rely on a ninety-
year-old Supreme Court case for the proposition that
“[p]unctuation is a minor, and not a controlling, element in
interpretation.” Barrett v. Van Pelt, 268 U.S. 86, 91 (1925); see
also NACS, 958 F. Supp. 2d at 102 (calling Congress’s failure to
use commas a “red herring”). This decision provides the
merchants little help. Not only was it written long before the
development of modern approaches to statutory interpretation,
see U.S. National Bank of Oregon v. Independent Insurance
Agents of America, Inc., 508 U.S. 439, 454–55 (1993) (noting
that although reliance on punctuation must not “distort[] a
statute’s true meaning,” “[a] statute’s plain meaning must be
enforced, of course, and the meaning of a statute will typically
heed the commands of its punctuation”), but it addressed
statutory language that, unlike here, contained a clearly
misplaced comma, Barrett, 268 U.S. at 88 (interpreting a statute
“so inapt and defective that it is difficult to give it a construction
that is wholly satisfactory” without ignoring its comma).

     The idea that we should entirely ignore punctuation would
make English teachers cringe. Even if punctuation is sometimes
a minor element in interpreting the meaning of language,
punctuation is often crucial—a reader might appropriately gloss
over a comma mistakenly inserted between a noun and a verb
yet pay extra attention to a comma or semicolon setting off
separate items in a list. Following the merchants’ advice and
stuffing punctuation to the bottom of the interpretive toolbox
would run the risk of distorting the meaning of statutory
language. After all, Congress communicates through written
language, and one component of written language is grammar,
including punctuation. As Strunk and White puts it, “the best
writers sometimes disregard the rules of rhetoric. When they do
                                21
so, however, the reader will usually find in the sentence some
compensating merit, attained at the cost of the violation. Unless
he is certain of doing as well, he will probably do best to follow
the rules.” WILLIAM STRUNK, JR. & E.B. WHITE, THE ELEMENTS
OF STYLE xvii–xviii (4th ed. 2000) (internal quotation marks
omitted). Put another way, “all our thoughts can be rendered
with absolute clarity if we bother to put the right dots and
squiggles between the words in the right places.” LYNN TRUSS,
EATS, SHOOTS & LEAVES 201–02 (2003).

     In this instance, the absence of commas matters far more
than Congress’s use of the word “which” rather than “that.”
Widely-respected style guides expressly require that commas set
off descriptive clauses, but refer to descriptive “which” and
restrictive “that” as a style preference rather than an ironclad
grammatical rule. As The Chicago Manual of Style explains:
    A relative clause that is restrictive—that is, essential to
    the meaning of the sentence—is neither preceded nor
    followed by a comma. But a relative clause that could
    be omitted without essential loss of meaning (a
    nonrestrictive clause) should be both preceded and (if
    the sentence continues) followed by a comma.
    Although which can be used restrictively, many careful
    writers preserve the distinction between restrictive that
    (no commas) and descriptive which (commas).
THE CHICAGO MANUAL OF STYLE 250 (14th ed. 2003). Compare
STRUNK & WHITE at 3–4 (“Nonrestrictive relative clauses are
parenthetic. . . . Commas are therefore needed.”), and WILSON
FOLLETT, MODERN AMERICAN USAGE: A GUIDE 69 (Erik
Wensberg ed., 1998) (same), with STRUNK & WHITE at 59 (“The
use of which for that is common in written and spoken language.
. . . Occasionally which seems preferable to that . . . But it would
be a convenience to all if these two pronouns were used with
                                22
precision.”), and FOLLETT at 293 (“The alert reader will notice
that quite a few excellent authors decline to use that and which
in precisely the ways that late-twentieth-century grammar books
recommend.”).

     In fact, elsewhere in the Durbin Amendment Congress
demonstrated that it is among those writers who ignore the
distinction between descriptive “which” and restrictive “that.” In
section 920(b)(1)(A), for example, Congress instructed the
Board to prevent networks and issuers from activating on a debit
card only one network or “2 or more such networks which are
owned, controlled, or otherwise operated by” the same company.
15 U.S.C. § 1693o-2(b)(1)(A)(i)-(ii) (emphasis added). Even
though Congress used the word “which” to introduce this clause,
the clause is clearly restrictive. A descriptive reading would
require that the Board prevent issuers and networks from ever
activating “one network” or “2 or more such networks.” In other
words, a descriptive reading would prevent the activation of any
networks at all, rendering debit cards useless chunks of plastic.
Cf. Barnhart v. Thomas, 540 U.S. 20, 24 (2003) (finding a
restrictive clause in the statutory phrase “any other kind of
substantial gainful work which exists in the national economy”).
By contrast, in the Durbin Amendment Congress set aside every
clearly descriptive clause with commas. See, e.g., 15 U.S.C. §
1693o-2(a)(4)(B)(ii) (“other costs incurred by an issuer which
are not specific to a particular electronic debit transaction, which
costs shall not be considered under paragraph (2)” (emphasis
added)).

     The merchants also emphasize Congress’s use of the terms
“distinguish between,” 15 U.S.C. § 1693o-2(a)(4)(B), and “other
costs,” id. § 1693o-2(a)(4)(B)(ii). According to the merchants,
the term “distinguish between” suggests that Congress required
the Board to “differentiate [between] the two categories of
                               23
costs,” and “the very use of the term ‘other costs’—as opposed
to simply ‘costs’—indicates the entire universe of costs that is
remaining after consideration of includable costs.” Appellees’
Br. 28. As noted above, these terms might provide some textual
support for the merchants’ preferred reading of the statute. But
given the Board’s reasonable determination that issuers incur
costs, other than incremental ACS costs, that are “specific to a
particular transaction,” the terms “distinguish between” and
“other costs” hardly compel the conclusion that the Board must
interpret section 920(a)(4)(B) as encompassing all costs that
issuers incur. Imagine that you make a deal to hand over part of
your baseball card collection and to distinguish between rookie
cards, which you must hand over, and other cards less than five
years old, which you must not. Although it would probably
make little financial sense, you could certainly hand over a 1960
Harmon Killebrew Topps card without violating the terms of the
deal.

     Next, the merchants assert that the Board, by inferring the
existence of a third category of costs, improperly reads a
delegation of authority into congressional silence. According to
the merchants, “Congress would not delineate with specificity
the characteristics of includable costs (e.g., incremental) if it
intended, by its silence, to allow the Board to consider and
include their opposite (e.g., nonincremental).” Appellees’ Br.
31; accord American Petroleum Institute v. Environmental
Protection Agency, 198 F.3d 275, 278 (D.C. Cir. 2000) (“[I]f
Congress makes an explicit provision for apples, oranges and
bananas, it is most unlikely to have meant grapefruit.”). But
section 920(a)(3)(A) clearly grants the Board authority to
promulgate regulations ensuring that interchange fees are
reasonable and proportional to costs issuers incur. The question
then is how section 920(a)(4)(B) limits the Board’s discretion to
define the statutory term “cost incurred by the issuer with
                                24
respect to the transaction,” not whether that section affirmatively
grants the Board authority to allow issuers to recover certain
costs.

     Finally, in a footnote the merchants point to section
920(a)(3)(B)’s requirement that the Board disclose certain ACS
cost information and to section 920(a)(4)(A)’s requirement that
the Board “consider the functional similarity between electronic
debit transactions and checking transactions that are required
within the Federal Reserve bank system to clear at par.” The
district court relied heavily on these provisions, concluding that
Congress’s decisions to limit disclosure “to the same costs
specified in section (a)(4)(B)(i)” and to direct the Board to
consider similarities, but not differences, between checks and
debit cards support the merchants’ interpretation of the statute.
NACS, 958 F. Supp. 2d at 103–04. But even assuming the
disclosure provision mirrors section 920(a)(4)(B)(i)’s reference
to incremental ACS costs—the word “incremental” appears
nowhere in the disclosure provision—the statute also allows the
Board to collect “such information as may be necessary to carry
out the provisions of this section,” not just information about
incremental ACS costs. 15 U.S.C. § 1693o-2(a)(3)(B). Similarly,
Congress’s instruction to the Board to “consider the functional
similarity between electronic debit transactions and checking
transactions” hardly precludes the Board from considering
differences as well. Doing just that, the Board decided that it
could allow banks to recover some costs in the debit card
context that they are unable to recover in the checking context.

     Given the Durbin Amendment’s ambiguity as to the
existence of a third category of costs, we must defer to the
Board’s reasonable determination that the statute splits costs into
three categories: (1) incremental ACS costs, which the Board
must allow issuers to recover; (2) costs specific to a particular
                                 25
transaction, other than incremental ACS costs, which the Board
may, but need not, allow issuers to recover; and (3) costs not
specific to a particular transaction, which the Board may not
allow issuers to recover. See Chevron, 467 U.S. at 843
(“Sometimes the legislative delegation to an agency on a
particular question is implicit rather than explicit. In such a case,
a court may not substitute its own construction of a statutory
provision for a reasonable interpretation made by the
administrator of an agency.”).

                                 B.
     Because the Board reasonably interpreted the Durbin
Amendment as allowing issuers to recover some costs in
addition to incremental ACS costs, we must now determine
whether the Board reasonably concluded that issuers can recover
the four specific types of costs the merchants challenge: “fixed”
ACS costs, network processing fees, fraud losses, and
transactions-monitoring costs. Much like agency ratemaking,
determining whether issuers or merchants should bear certain
costs is “far from an exact science and involves policy
determinations in which the [Board] is acknowledged to have
expertise.” Time Warner Entertainment Co. v. Federal
Communications Commission, 56 F.3d 151, 163 (D.C. Cir.
1995) (internal quotation marks omitted). We afford agencies
special deference when they make these sorts of determinations.
See, e.g., BNSF Railway Co. v. Surface Transportation Board,
526 F.3d 770, 774 (D.C. Cir. 2008) (“In the rate-making area,
our review is particularly deferential, as the Board is the expert
body Congress has designated to weigh the many factors at issue
when assessing whether a rate is just and reasonable.”); Time
Warner, 56 F.3d at 163. With that caution in mind, we address
each category of costs.
                                26
                       “Fixed” ACS Costs

     Microeconomics textbooks draw a clear distinction between
“fixed” and “variable” costs: fixed costs are incurred regardless
of transaction volume, whereas variable costs change as
transaction volume increases. E.g., N. GREGORY MANKIW,
PRINCIPLES OF MICROECONOMICS 276–77 (3d ed. 2004). The
merchants, noting that the statute precludes recovery of costs
“not specific to a particular . . . transaction,” 15 U.S.C. § 1693o-
2(a)(4)(B)(ii), argue that the Board’s Final Rule improperly
allows recovery of fixed costs such as “equipment, hardware and
software.” Appellees’ Br. 35. “By definition,” the merchants
declare, “fixed costs are not ‘specific’ to any ‘particular’
transaction and fall squarely within the statute’s excludable costs
provision.” Id. at 39. The merchants therefore urge us to require
the Board to return to something along the lines of its proposed
rule, under which merchants could only recover average variable
ACS costs.

     The merchants’ argument certainly has some persuasive
power. One might think it a stretch if a shoe store claimed that
the rent it pays its landlord is somehow “specific” to a
“particular” shoe sale. But the merchants have never argued that
issuers should be allowed to recover only costs incurred as a
result of processing individual, isolated transactions. See NPRM,
75 Fed. Reg. at 81,736 (requesting comment about whether
“costs should be limited to the marginal cost of a transaction”);
Final Rule, 76 Fed. Reg. at 43,427 n.118 (noting that “[t]he
Board did not receive comments regarding the use of marginal
cost”). Indeed, the Board’s proposed rule, which the merchants
seem to endorse, would have allowed recovery of costs that are
variable over the course of a year but could not be traced to any
one particular transaction.
                                27
     We think the Board reasonably declined to read section
920(a)(4)(B) as preventing issuers from recovering “fixed”
costs. As the Board pointed out, the distinction the merchants
urge between what they refer to as non-includable “fixed” costs
and includable “variable” costs depends entirely on whether, on
an issuer-by-issuer basis, certain costs happen to vary based on
transaction volume in a particular year. For example, in any
given year one issuer might classify labor as an includable cost
because labor costs happened to vary based on transaction
volume over that year, while another issuer might classify labor
as a non-includable cost because such costs happened to remain
fixed over that year. See Final Rule, 76 Fed. Reg. at 43,427.
Moreover, the Board pointed out, the distinction between
variable and fixed ACS costs depends in some instances on
whether an issuer “performs its transactions processing in-
house” or “outsource[s] its debit card operations to a third-party
processor that charge[s] issuers a per-transaction fee based on its
entire cost.” Id. In any event, the Board concluded, requiring
issuers to segregate includable “variable” costs from excludable
“fixed” costs on a year-by-year basis would prove “exceedingly
difficult for issuers . . . [because] even if a clear line could be
drawn between an issuer’s costs that are variable and those that
are fixed, issuers’ cost-accounting systems are not generally set
up to differentiate between fixed and variable costs.” Id. The
Board therefore determined that any distinction between fixed
and variable costs would prove artificial and unworkable.

     Instead, pointing out that the statute requires interchange
fees to be “reasonable and proportional” to issuer costs, the
Board interpreted section 920(a)(4)(B) as allowing issuers to
recover costs they must incur in order to effectuate particular
electronic debit card transactions but precluding them from
recovering other costs too remote from the processing of actual
transactions. “This reading interpret[s] costs that ‘are not
                                28
specific to a particular electronic debit transaction,’ and . . .
cannot be considered by the Board, to mean those costs that are
not incurred in the course of effecting any electronic debit
transaction.” Id. at 43,426. In our view, the Board reasonably
distinguished between costs issuers could recover and those they
could not recover on the basis of whether those costs are
“incurred in the course of effecting” transactions. Id. For
instance, the Board’s rule allows issuers to recover equipment,
hardware, software, and labor costs since “[e]ach transaction
uses the equipment, hardware, software and associated labor,
and no particular transaction can occur without incurring these
costs.” Id. at 43,430. By contrast, the rule precludes issuers from
recovering the costs of producing and distributing debit cards
because “an issuer’s card production and delivery costs . . . are
incurred without regard to whether, how often, or in what way
an electronic debit transaction will occur.” Id. at 43,428. Given
the Board’s expertise, we see no basis for upsetting its
reasonable line-drawing. See ExxonMobil Gas Marketing Co. v.
Federal Energy Regulatory Commission, 297 F.3d 1071, 1085
(D.C. Cir. 2002) (“We are generally unwilling to review line-
drawing . . . unless a petitioner can demonstrate that lines drawn
. . . are patently unreasonable, having no relationship to the
underlying regulatory problem.” (internal quotation marks
omitted)).

                   Network Processing Fees

     This is easy. Network processing fees, which issuers pay on
a per-transaction basis, are obviously specific to particular
transactions. The merchants argue that allowing issuers to
recover network processing fees through the interchange fee
would run afoul of section 920(a)(8)(B), which requires the
Board to ensure that “a network fee is not used to directly or
indirectly compensate an issuer with respect to an electronic
                                29
debit transaction.” Perhaps signaling that even the merchants are
not entirely confident about this argument, they present it only in
a footnote. The merchants should have left it out entirely. As the
Board points out, section 920(a)(8)(B) is designed to prevent
issuers and networks from circumventing the Board’s
interchange fee rules, not to prevent issuers from recovering
reasonable network processing fees through the interchange fee.
Final Rule, 76 Fed. Reg. at 43,442 (“[Section 920(a)(8)(B)]
authorizes the Board to prescribe rules to prevent circumvention
or evasion of the interchange transaction fee standards.”).

                          Fraud Losses

     The merchants nowhere challenge the Board’s conclusion
that fraud losses, which result from the settlement of particular
fraudulent transactions, are specific to those transactions. The
only question is whether a separate provision of the Durbin
Amendment—section 920(a)(5)’s fraud-prevention adjustment,
which allows issuers to recover fraud-prevention costs if those
issuers comply with the Board’s fraud-prevention standards—
precludes the Board from allowing issuers to recover fraud
losses as part of section 920(a)(2)’s “reasonable and
proportional” interchange fee. The merchants claim that it does.

     First, noting that Congress intended the fraud-prevention
adjustment to be the only “fraud-related adjustment of the
issuer,” 15 U.S.C. § 1693o-2(a)(5)(A)(ii)(I), the merchants argue
that the Board should have allowed issuers to recover fraud-
related costs only through the fraud-prevention adjustment. We
disagree. The Board determined—reasonably in our view—that
because fraud losses result from the failure of fraud-prevention,
they do not themselves qualify as fraud-prevention costs. See
Final Rule, 76 Fed. Reg. at 43,431 (“An issuer may experience
losses for fraud that it cannot prevent and cannot charge back to
the acquirer or recoup from the cardholder.”). And nothing in the
                               30
statute suggests that Congress used the word “adjustment” to
describe the process of determining which costs issuers should
be allowed to recover directly through the interchange fee.
Rather, when discussing the fraud-prevention adjustment,
Congress empowered the Board to “allow for an adjustment to
the fee amount received or charged by an issuer under paragraph
(2).” 15 U.S.C. § 1693o-2(a)(5)(A). Paragraph (2), in turn,
requires that the interchange fee be “reasonable and
proportional” to costs incurred by issuers. Id. § 1693o-2(a)(2).
Thus, Congress used the word “adjustment” to describe a bonus
over and above the “reasonable and proportional” interchange
fee.

     The merchants next maintain that allowing issuers to
recover fraud losses through the interchange fee “irrespective of
any particular bank’s efforts to reduce fraud” would undermine
Congress’s decision to condition receipt of the fraud-prevention
adjustment on compliance with the Board’s fraud-prevention
standards. Appellees’ Br. 43. Even assuming the merchants’
policy argument has some merit—allowing recovery of fraud
losses regardless of compliance with fraud-prevention standards
might well decrease issuers’ incentives to invest in fraud
prevention—the Board rejected it, reasoning that “[i]ssuers will
continue to bear the cost of some fraud losses and cardholders
will continue to demand protection against fraud.” Final Rule,
76 Fed. Reg. at 43,431. Such policy judgments are the province
of the Board, not this Court. See Village of Barrington, Illinois
v. Surface Transportation Board, 636 F.3d 650, 666 (D.C. Cir.
2011) (“As long as the agency stays within [Congress’s]
delegation, it is free to make policy choices in interpreting the
statute, and such interpretations are entitled to deference.”
(internal quotation marks omitted) (alterations in original)).
                               31
                Transactions-Monitoring Costs

     The Board acknowledged in the Final Rule that
transactions-monitoring costs, unlike fraud losses, are the
paradigmatic example of fraud-prevention costs. Final Rule, 76
Fed. Reg. at 43,397 (“The most commonly reported fraud
prevention activity was transaction monitoring.”). The Board
then distinguished between “[t]ransactions monitoring systems
[that] assist in the authorization process by providing
information to the issuer before the issuer decides to approve or
decline the transaction,” which the Board placed outside the
fraud-prevention adjustment, and “fraud-prevention activities . .
. that prevent fraud with respect to transactions at times other
than when the issuer is effecting the transaction”—for instance
the cost of sending “cardholder alerts . . . inquir[ing] about
suspicious activity”—which the Board determined should be
“considered in connection with the fraud-prevention
adjustment.” Id. at 43,430–31. Challenging this distinction, the
merchants think it “preposterous to suggest that Congress would
specifically address the costs associated with fraud prevention in
a separate provision of the statute, condition the recovery of
those costs on an issuer’s compliance with fraud prevention
measures, and then . . . permit recovery of those very same
costs” whether or not an issuer complies with fraud-prevention
standards. Appellees’ Br. 41.

     As an initial matter, we agree with the Board that
transactions-monitoring costs can reasonably qualify both as
costs “specific to a particular . . . transaction” (section
920(a)(4)(B)) and as fraud-prevention costs (section 920(a)(5)).
Thus, the Board may have discretion either to allow issuers to
recover transactions-monitoring costs through the interchange
fee regardless of compliance with fraud-prevention standards or
to preclude issuers from recovering transactions-monitoring
                                32
costs unless those issuers comply with fraud-prevention
standards. That said, “an agency must cogently explain why it
has exercised its discretion in a given manner.” Motor Vehicle
Manufacturers Association of the United States v. State Farm
Mutual Automobile Insurance Co., 463 U.S. 29, 48 (1983). We
agree with the merchants that the Board has fallen short of that
standard.

     The Board insists that the distinction it drew between fraud-
prevention costs falling outside the fraud-prevention adjustment
and fraud-prevention costs falling within it reflects the
distinction between, on the one hand, section 920(a)(4)(B)’s
focus on a single transaction and, on the other, section
920(a)(5)(A)(i)’s focus on “electronic debit transactions
involving that issuer.” According to the Board, Congress
“intended the . . . fraud-prevention adjustment to take into
account an issuer’s fraud prevention costs over a broad spectrum
of transactions that are not linked to a particular transaction.”
Appellant’s Br. 66–67. But as noted above, the Board
interpreted the term “specific to a particular . . . transaction” as
in fact allowing recovery of many costs not literally “specific” to
any one “particular” transaction. See supra at 26–28. The costs
of hardware, software, and labor seem no more “specific” to one
“particular” transaction than many of the fraud-prevention costs
the Board determined fall within the fraud prevention
adjustment. The Board’s own interpretation of the statute thus
undermines its justification for concluding that Congress
established a fraud-prevention adjustment, conditioned receipt of
that adjustment on compliance with fraud-prevention standards,
yet allowed issuers to recover the paradigmatic example of
fraud-prevention costs—transactions-monitoring costs—whether
or not issuers comply with those standards.
                                33
     All that said, the Board may well be able to articulate a
reasonable justification for determining that transactions-
monitoring costs properly fall outside the fraud-prevention
adjustment. But the Board has yet to do so. “If the record before
the agency does not support the agency action, if the agency has
not considered all relevant factors, or if the reviewing court
simply cannot evaluate the challenged agency action on the
basis of the record before it, the proper course, except in rare
circumstances, is to remand to the agency for additional
investigation or explanation.” Florida Power & Light Co. v.
Lorion, 470 U.S. 729, 744 (1985) (emphasis added). We shall do
so here. Because the interchange fee rule generally rests on a
reasonable interpretation of the statute, because the Board may
well be able to articulate a sufficient explanation for its
treatment of fraud-prevention costs, and because vacatur of the
rule would be disruptive—the merchants seek an even lower
interchange fee cap, but vacating the Board’s rule would lead to
an entirely unregulated market, allowing the average interchange
fee to once again reach or exceed 44 cents per transaction—we
see no need to vacate. See Heartland Regional Medical Center
v. Sebelius, 566 F.3d 193, 198 (D.C. Cir. 2009) (noting that
remand without vacatur is warranted “[w]hen an agency may be
able readily to cure a defect in its explanation of a decision” and
the “disruptive effect of vacatur” is high); see also, e.g.,
Environmental Defense Fund v. Environmental Protection
Agency, 898 F.2d 183, 190 (D.C. Cir. 1990) (instructing that
courts should ordinarily remand without vacatur when vacatur
would “at least temporarily defeat” the interests of the party
successfully seeking remand).

                               III.
     Having resolved the merchants’ challenges to the
interchange fee rule, we turn to the anti-exclusivity rule. As
explained above, see supra at 9, section 920(b) requires the
                               34
Board to promulgate regulations preventing “an issuer or
payment card network” from “restrict[ing] the number of
payment card networks on which an electronic debit transaction
may be processed” to a single network, or to networks affiliated
with one another. In the proposed rule, the Board outlined two
alternatives: require issuers and networks to activate two
unaffiliated networks or two unaffiliated networks for each
method of authentication. In the Final Rule, the Board chose the
former, requiring activation of two unaffiliated networks on each
debit card regardless of method of authentication.

     The merchants believe that the Durbin Amendment
unambiguously requires that all merchants have multiple
unaffiliated network routing options for each debit transaction.
See NACS, 958 F. Supp. 2d at 109–12 (accepting this argument).
Arguing that the Board’s rule flunks this requirement, the
merchants emphasize two undisputed facts. First, given that
most merchants refuse to accept PIN debit, many transactions
can currently be processed only on signature debit. Second,
cardholders, not merchants, often have the ability to select
whether to process transactions on signature networks or PIN
networks. As a result, the merchants emphasize, under the
Board’s rule many merchants will still lack the ability to choose
between unaffiliated networks when deciding how to process
particular transactions. Disputing none of this, the Board points
out that all merchants could accept PIN debit even if some
choose not to and emphasizes that the statute is silent about
“restrictions imposed by merchants or consumers that limit
routing choice.” Appellant’s Br. 22. Given the parties’
agreement that under the Board’s rule some merchants will lack
routing choice for particular transactions, we must determine
whether the statute requires that all merchants—even those who
voluntarily choose not to accept PIN debit—have the ability to
decide between unaffiliated networks when routing transactions.
                                35
     The merchants have a steep hill to climb. Congress directed
the Board to issue rules that would accomplish a particular
objective, leaving it to the Board to decide how best to do so,
and the Board’s rule seems to comply perfectly with Congress’s
command. Under the rule, “issuer[s] and payment card
network[s]” cannot “restrict the number of payment card
networks on which an electronic debit transaction may be
processed” to only affiliated networks—exactly what the statute
requires. 15 U.S.C. § 1693o-2(b)(1)(A).

    Undaunted, the merchants emphasize one largely
conclusory textual argument and allude to another. First, relying
on the statutory phrase “electronic debit transaction,” id. §
1693o-2(b)(1)(A), they maintain that the statute plainly “requires
the Board to ensure that merchants be afforded a choice of
networks for each debit transaction.” Appellees’ Br. 45. But
context matters. Relying on the statute’s reference to “issuer[s]
and payment card network[s],” the Board reasonably read the
“electronic debit transactions” phrase to prevent issuers and
networks, prior to instigation of any particular debit transaction,
from limiting the number of networks “on which an electronic
debit transaction may be processed” to only affiliated networks.
15 U.S.C. § 1693o-2(b)(1)(A) (emphasis added).

     In a footnote, the merchants repeat, though they seem not to
embrace, a textual argument on which the district court relied.
Looking to the statutory definitions of “electronic debit
transaction” (“a transaction in which a person uses a debit card”)
and of “debit card” (“any card . . . issued or approved for use
through a payment card network to debit an asset account . . .
whether authorization is based on signature, PIN, or other
means”), id. § 1693o-2(c)(2), (c)(5), the district court ruled that
the statutory term “electronic debit transaction” requires that
issuers and networks activate multiple unaffiliated networks for
                               36
each transaction “whether authorization is based on signature,
PIN, or other means,” NACS, 958 F. Supp. 2d at 110–11. But we
think it quite implausible that Congress engaged in a high-stakes
game of hide-and-seek with the Board, writing a provision that
seems to require one thing but embedding a substantially
different and, according to financial services amici, much more
costly requirement in the statute’s definitions section. Cf.
Whitman v. American Trucking Association, 531 U.S. 457, 468
(2001) (“Congress . . . does not . . . hide elephants in
mouseholes.”).

     The merchants also argue that the Board’s rule runs afoul of
the Durbin Amendment’s purpose. Pointing out that Congress
intended network competition to drive down network processing
fees, the merchants insist that the Board has undermined this
competitive market because “merchants will be deprived of
network choice for a substantial segment of debit transactions in
the marketplace today.” Appellees’ Br. 47. But the Board
thought differently. As it explained in the Final Rule,
“merchants that currently accept PIN debit would have routing
choice with respect to PIN debit transactions in many cases
where an issuer chooses to participate in multiple PIN debit
networks.” Final Rule, 76 Fed. Reg. at 43,448. Indeed, the Board
presents uncontested evidence demonstrating that its rule has, as
predicted, substantially increased network competition.
According to the Board, as a result of the rule over 100 million
debit cards were activated on new networks, and “[Visa], which
had previously accounted for approximately 50-60% of the [PIN
debit] market, lost roughly half that share.” Appellant’s Br. 37 &
n.6 (internal quotation marks omitted).

    Of course, as the Board acknowledges, the merchants’
preferred rule would result in more competition. But in its Final
Rule the Board explained the policy considerations that led it to
                                37
reject that approach. For one thing, cardholders might prefer to
route transactions over certain networks, perhaps because they
believe those networks to have better fraud-prevention policies.
Final Rule, 76 Fed. Reg. at 43,447–48. Also, the merchants’
preferred rule “could potentially limit the development and
introduction of new authentication methods” since issuers would
be unable to compel merchants to accept new authentication
techniques. Final Rule, 76 Fed. Reg. at 43,448. The merchants
ignore these reasonable concerns. Given that the Board’s rule
advances the Durbin Amendment’s purpose, we decline to
second-guess its reasoned decision to reject an alternative option
that might have further advanced that purpose.

     Next, the merchants emphasize the interaction between
section 920(b)’s two key components: the anti-exclusivity and
routing priority provisions. According to the merchants, the
Board’s anti-exclusivity rule renders the routing priority
provision meaningless, since merchants will often lack the
ability to choose between multiple unaffiliated routing options.
But as the Board points out, the merchants misunderstand the
routing priority provision. Recall that it prohibits issuers and
networks from requiring merchants to process transactions over
certain activated networks rather than others. Far from rendering
the routing priority provision a nullity, the Board’s anti-
exclusivity provision would be ineffective without it. Absent the
routing priority provision, issuers and networks could, for
instance, activate two PIN networks and a signature network
affiliated with one of the PIN networks and then require
merchants to route transactions over the PIN network affiliated
with the signature network rather than over the other PIN
network.

     Finally, the merchants question the Board’s premise that it
is they, not issuers and networks, who restrict routing options for
                               38
transactions under the Board’s Final Rule. To this end, they
assert that issuer and network rules arbitrarily prevent merchants
from processing PIN transactions on signature networks and vice
versa, suggesting that the Board could comply with the statute
by eliminating the distinction between PIN and signature debit.
But even if issuers and networks are responsible for maintaining
this distinction—a point they strongly dispute—merchants, not
issuers or networks, limit their own options when they refuse to
accept PIN debit, and cardholders, not issuers or networks, limit
merchants’ options when given the ability to choose how to
process transactions. “The principal fallacy with the Merchants’
argument,” the Board aptly explains, “is that they selectively
view transactions only from their own perspective and only after
the point at which the merchant itself or the consumer may have
elected to restrict certain routing options,” whereas “section
920(b) speaks only in terms of issuer and payment card network
restrictions” imposed prior to initiation of any particular debit
card transaction. Reply Br. 2–3.

    In sum, far from summiting the steep hill, the merchants
have barely left basecamp. We therefore defer to the Board’s
reasonable interpretation of section 920(b) and reject the
merchants’ challenges to the anti-exclusivity rule.

                               IV.
     For the foregoing reasons, we reverse the district court’s
grant of summary judgment to the merchants and remand for
further proceedings consistent with this opinion.

                                                     So ordered.