Regulation AT's 2015 plan put algorithm controls at three layers; the CFTC's 2021 rule kept the exchange's
On November 24, 2015 the CFTC approved Regulation AT, a proposal requiring risk controls, development standards and a source code repository at trading firms, plus controls at clearing firms and exchanges. It was withdrawn in 2020; 2021 risk principles put the duty on exchanges.
Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.
Two-minute summaryThe dek and the first sentence of each section, quoted from the article
On November 24, 2015 the CFTC approved Regulation AT, a proposal requiring risk controls, development standards and a source code repository at trading firms, plus controls at clearing firms and exchanges. It was withdrawn in 2020; 2021 risk principles put the duty on exchanges.
- The week Regulation AT drew opposition over source code, among other provisions. [5] Its more lasting contribution was architectural.
- The perimeter was drawn around automation, not speed The exclusion was narrow. [2] An order was outside the definition only if every parameter was manually entered by a natural person, with no further discretion by any computer system or algorithm. [2] In our reading, that would have reached common tools that automate a single decision, such as an order that a system splits into smaller parts or times for submission.
- The three layers each carried their own controls The clearing firm's compliance report would describe its program for those controls, in the aggregate. [2] The chief executive officer or chief compliance officer would certify that, to the best of their knowledge and reasonable belief, the report was accurate and complete. [2] In our reading, that certification put personal accountability at the clearing layer for controls that ran, in part, on exchange systems.
- The engineering lifecycle became a regulated process It said the Commission believed the risk controls required of trading firms and clearing firms were already widely used by market participants. [3] It expected upgrades to impose one-time costs but generally not increased annual costs. [3] In our reading, that assumption cut both ways.
- The source code record was where the proposal broke The repository would manage source code access, persistence, copies of all code used in production, and changes to that code. [3] Its audit trail would show, for each material change, who made it, when, and the coding purpose of the change. [3] In our reading, this is an ordinary engineering record in any well-run development shop.
- What survived was the exchange's control Set side by side, the two designs differ in where the duty sits.
The week
On November 24, 2015, the Commodity Futures Trading Commission announced that it had unanimously approved proposed rules on automated trading. [1] The press release called the proposal a comprehensive regulatory response to the evolution of automated trading on U.S. designated contract markets, the futures exchanges the CFTC regulates. [1] The rules were known collectively as Regulation Automated Trading, or Regulation AT. [1]
The release described a multilevel approach. [1] The proposal set risk control and other requirements for three groups. [1] They were market participants using algorithmic trading systems, defined as "AT Persons"; clearing member futures commission merchants with respect to their AT Person customers; and designated contract markets executing AT Person orders. [1]
A fact sheet issued the same day set out the definitions. [2] "Algorithmic Trading" would mean trading on or subject to the rules of an exchange where computer algorithms or systems determine whether to initiate, modify or cancel an order, or make other determinations about it. [2] The definition listed determinations such as the product, venue, type, timing, price and quantity of an order. [2]
The definition excluded orders whose every parameter was manually entered into a front-end system by a natural person, with no further discretion by any computer system or algorithm. [2] An "AT Person" would be a registrant, such as a futures commission merchant, swap dealer, commodity pool operator or commodity trading advisor, that engages in Algorithmic Trading on an exchange. [2] It would also include a new class of persons required to register as floor traders. [2]
That new class came from a registration proposal. [1] [2] The release said Regulation AT proposed to require registration of certain proprietary traders that, while responsible for significant trading volumes in key futures products, were not registered with the Commission. [1] The fact sheet said the requirement would apply to proprietary traders engaged in algorithmic trading through direct electronic access to an exchange. [2]
For AT Persons, the fact sheet listed three key elements. [2] They would implement pre-trade risk controls, including limits on order message and execution frequency per unit time, order price and maximum order size parameters, and order cancellation systems. [2] They would implement standards for development, testing and monitoring of their systems. [2] They would submit annual compliance reports to exchanges and keep books and records for inspection by exchanges. [2]
The development standards named specific practices. [2] They included keeping the development environment separate from production, testing before implementation, a source code repository, and real-time monitoring of the systems. [2]
Clearing member futures commission merchants would implement risk controls for algorithmic orders from AT Persons. [2] For orders sent through direct electronic access, they would implement exchange-provided controls; for other orders, they would set the controls themselves. [2] They would submit compliance reports to exchanges, certified by the chief executive officer or chief compliance officer. [2]
Exchanges would implement pre-trade risk controls and order cancellation systems for algorithmic orders, with parallel controls for manually submitted orders. [2] They would require compliance reports from AT Persons and clearing firms, review them, identify outliers and provide instructions for remediation. [2] They would provide test environments and self-trade prevention tools. [2]
The proposal also covered transparency. [1] [2] Exchanges would publish a description of rules or known attributes of their trade matching platforms that materially affect the time, price or quantity of execution. [1] They would make disclosures about market maker and trading incentive programs. [1] [2] All AT Persons would have to become members of a registered futures association, which would consider membership rules on algorithmic trading. [1] [2]
The release said the notice would be open for a 90-day public comment period. [1] The fact sheet said it would be published in the Federal Register. [2]
Regulation AT drew opposition over source code, among other provisions. [5] Its more lasting contribution was architectural. It asked where the controls on an automated trading system should sit, and answered: everywhere. [2] The trading firm, the clearing firm and the exchange would each run pre-trade controls, and the firm would also run an engineering discipline the Commission could inspect. [2] [3]
Our argument is that the proposal's fate turned on that placement. The CFTC withdrew it in 2020 and rejected three of its approaches: new registration, source code access without a subpoena, and prescriptive control lists across exchanges, clearing firms and others. [5] In 2021 it adopted principles that put the duty on the exchange. [6] In our reading, the control that survived was the one the regulator could already reach.
For leaders at banks and broker-dealers that run trading or clearing businesses, the episode is a lesson in control architecture. A control placed at a layer the regulator does not directly supervise needs a new legal hook, such as registration or record access. Those hooks were among the provisions the 2020 withdrawal rejected by name. [5]
We take it in five parts, following an order from the algorithm to the market:
- The perimeter: which systems the definition would have caught.
- The three layers: what each of the firm, the clearing broker and the exchange would have done.
- The engineering lifecycle: development, testing, monitoring and the people who run them.
- The record: source code as a regulated book, and why that broke.
- What survived: exchange-based controls under principles, and what firms do now.
The perimeter was drawn around automation, not speed
The proposal's definition of Algorithmic Trading did not turn on how fast a firm traded. [2] It turned on whether a computer made determinations about an order. [2] Any order where an algorithm chose the product, the timing, the price, the quantity or how to split it would be in. [2]
The manual-order exclusion
The exclusion was narrow. [2] An order was outside the definition only if every parameter was manually entered by a natural person, with no further discretion by any computer system or algorithm. [2] In our reading, that would have reached common tools that automate a single decision, such as an order that a system splits into smaller parts or times for submission.
The proposal's own estimate suggests the scale. [3] Using a sample of orders, the Commission estimated about 350 potential AT Persons, and increased that figure by 20 percent to 420 for its calculations. [3] It said some firms in the count might not meet the definition, possibly making the estimate higher than the actual number, and also that the estimate could be incomplete. [3]
Registration as the hook
The perimeter mattered because of who was inside it. [2] An AT Person had to be a registrant, and the proposal would have created registration for proprietary traders using direct electronic access who were not otherwise registered. [1] [2] The proposal estimated approximately 100 potential firms that might need to register. [3]
In our reading, registration was how the Commission would reach firms whose orders it otherwise saw only through brokers and exchanges. That made it essential to the design. It also made registration a target for opponents: the 2020 withdrawal recorded commenters' opposition to identifying and registering AT Persons, and rejected new registration by name. [5]
Direct electronic access
The proposal gave special weight to one access method. [2] The fact sheet defined direct electronic access as an arrangement where a person electronically sends an order to an exchange without the order first being routed through a separate clearing member. [2] Orders sent this way bypass the clearing firm's own systems. [2]
That is why the proposal treated these orders differently at the clearing layer. [2] A clearing firm could not run its own controls on an order it never saw, so it would implement controls the exchange provided. [2] In our reading, the definition marked the point where the clearing firm's visibility ended and the exchange's began.
What the perimeter means for a bank
A bank's trading and treasury desks can use execution algorithms supplied by brokers or vendors. Under the proposal, whether that made the bank an AT Person depended on its registration category and whether it engaged in Algorithmic Trading on a futures exchange. [2] A swap dealer or futures commission merchant using such tools on an exchange fit the definition. [2]
The proposal did not stay static on this point. [4] The 2016 supplemental proposal proposed modifications to the trading firm registration criteria and to compliance options for trading firms that use third-party algorithmic trading systems. [5] In our reading, that acknowledged a common arrangement: firms that run algorithms they did not write.
The volume threshold that came later
The Commission revisited the perimeter in 2016. [4] The supplemental proposal said it had considered comments that the original definitions might not successfully filter the scope of the rules. [4] It proposed adding a volume threshold test to the definition of AT Person. [4]
It explained the trade-off. [4] It acknowledged comments that any volume of trading could pose risks. [4] It said AT Person status carried compliance costs for risk controls, testing and recordkeeping, so it was appropriate to limit the population to larger market participants. [4] It said Regulation AT was not intended to capture large swaths of new or existing registrants. [4]
In our reading, the threshold changed the logic of the perimeter. The original definition caught a firm because a computer made its order decisions. [2] The threshold caught it because of how much it traded. [4] The first is a test of method; the second is a test of systemic weight. The 2021 principles abandoned both by applying exchange controls to all electronic orders. [6]
The three layers each carried their own controls
The proposal's distinctive feature was layering. [3] The Commission said commenters on its 2013 concept release had generally supported a multi-layered approach, with the trading firm, clearing firm and exchange each implementing controls adjusted to circumstance. [3]
The trading firm
AT Persons would implement pre-trade risk controls. [2] [3] The proposed rule text required at least maximum order message and execution frequency per unit time, and order price parameters and maximum order size limits. [3] The controls would be set at the level of each AT Person, or at a more granular level the firm chose, such as product or account. [3]
The firm would also need order cancellation systems. [3] These had to be able to disengage algorithmic trading immediately, cancel selected or all resting orders, and prevent new order messages. [3] Before its first algorithmic order on an exchange, the firm would notify the exchange whether its resting orders should be canceled or suspended if its system disconnected. [3]
The firm would monitor connectivity. [3] Firms with direct electronic access would implement systems indicating on an ongoing basis whether they had proper connectivity with the trading platform and the exchange's market data systems. [3] Natural person monitors would be promptly alerted when a pre-trade control was breached. [3]
The clearing broker
Clearing member futures commission merchants would implement risk controls for algorithmic orders originating with their AT Person customers. [2] For direct electronic access orders, which bypass the clearing firm's own systems, they would implement exchange-provided controls. [2] For other orders, they would set the controls themselves. [2]
The clearing firm's compliance report would describe its program for those controls, in the aggregate. [2] The chief executive officer or chief compliance officer would certify that, to the best of their knowledge and reasonable belief, the report was accurate and complete. [2] In our reading, that certification put personal accountability at the clearing layer for controls that ran, in part, on exchange systems.
The exchange
Exchanges would implement pre-trade risk controls and order cancellation systems for algorithmic orders, and parallel controls for manual orders. [2] They would require compliance reports from AT Persons and clearing firms, review them periodically, identify outliers and instruct remediation. [2] They would review books and records of AT Persons and clearing firms as necessary. [2]
In our reading, that gave exchanges a supervisory role over the compliance of trading and clearing firms, and the exchange would have become the place where compliance evidence from every layer came together.
Self-trade prevention
The exchange layer carried duties beyond risk controls. [2] Exchanges would establish self-trade prevention tools and either apply them or require participants to use them. [2] The fact sheet said self-trading would be defined as matching orders for accounts with common beneficial ownership or under common control. [2]
Exchanges could allow such matches where independent decision makers initiated the orders. [2] They would publish quarterly statistics disclosing approved self-trading. [2] In our reading, for a firm with several trading desks, this would have required an account map showing common ownership and control, and a record of which desks made decisions independently.
Disclosure of the matching engine
The proposal also reached the exchange's own technology. [1] [2] Exchanges would publicly describe rules, specifications or known attributes of their trade matching platforms that materially affect the time, priority, price or quantity of execution. [2] The disclosure would also cover the ability to cancel or modify orders and the transmission of market data and confirmations. [2]
In our reading, this treated the matching engine as infrastructure whose behavior participants were entitled to know. A firm designing an algorithm needs to understand how the venue prioritizes and confirms orders. The proposal would have made that a published disclosure rather than something learned by testing.
Incentive programs
Exchanges would also file market maker and trading incentive programs as rule filings, describing eligibility criteria and payments. [2] Payments would be prohibited for trades between accounts with common ownership. [2] In our reading, this linked the self-trade definition to the economics of incentive programs, closing a route by which a firm could earn rebates by trading with itself.
The clearing firm's position
The clearing firm stands behind its customers' trades. Its credit exposure grows with every order a customer's algorithm sends. The proposal's clearing-layer controls would have given the clearing firm a defined role in limiting that exposure for algorithmic orders. [2]
In our reading, the clearing firm gained something and lost something. It gained a regulatory basis for requiring controls from its AT Person customers, and for using exchange-provided controls on direct-access orders. [2] It took on certified compliance reports for controls partly outside its own systems. [2]
The redundancy problem
Layering creates overlap. The Commission recorded one commenter's concern, from the trade association FIA, that similar but not identically implemented controls at several points could add complexity and cause conflicting limits. [3] It noted the same commenter's view that some controls, such as maximum order size limits, could sit at both the firm and clearing levels without redundancy, because they reflected different responsibilities. [3]
The proposal tried to manage this with flexibility. [2] [3] It required particular controls while leaving the firm, clearing member or exchange discretion over design and parameters. [3] The Commission described this as a principles-based approach to its risk control requirements. [3]
Critics did not accept that description. The 2020 withdrawal recorded that commenters asserted the proposed rules were overly prescriptive and could result in redundant or overlapping risk controls. [5] It also recorded their view that the benefits were not commensurate with the costs. [5]
From three layers to two
The Commission changed the structure in 2016. [4] The supplemental proposal said that while some commenters supported the multi-layered approach, numerous commenters viewed it as unnecessarily redundant and prescriptive. [4] It proposed controls at two levels instead: the AT Person or the futures commission merchant, and the exchange. [4]
It explained the logic. [4] Two levels meant mistakes or omissions at one level would have a backstop. [4] Because a disruptive algorithm would affect other participants at the exchange level, the Commission would require exchange controls for all electronic orders, regardless of source. [4] The second set of controls would sit at the AT Person or the futures commission merchant, depending on who originated the order. [4]
In our reading, the 2016 change previewed the outcome. The one layer that appeared in every version of the design was the exchange's control over all electronic orders. [2] [4] [6] It was the layer the 2021 rule kept. [6]
What layering asks of a firm's architecture
For a firm, three layers of controls means three places an order can be stopped, and three sets of limits to reconcile. In our reading, a firm's architecture would need to know the exchange and clearing limits that applied to it, so its own controls tripped first and gave it the chance to act. Otherwise, the firm would learn of a problem from an exchange rejection.
That reconciliation is engineering work: a limit inventory, a mapping between the firm's own parameters and the exchange's, and alerts when one changes. The proposal did not require such an inventory. [3] It is the kind of artifact a firm would need to answer the compliance report the proposal did require. [2]
The engineering lifecycle became a regulated process
The proposal's development and testing standards read like a software engineering policy. [3] Each AT Person would implement written policies and procedures for developing and testing its algorithmic trading systems. [3]
What the standards required
The proposed rule text set out minimum elements. [3]
- A development environment adequately isolated from the production trading environment. [3]
- Testing of all algorithmic trading code and related systems, and any changes, before implementation, both internally and on each exchange where trading would occur. [3]
- Regular back-testing using historical transaction, order and message data. [3]
- Regular stress tests of systems under a variety of market conditions. [3]
- Procedures for documenting the strategy and design of proprietary algorithmic trading software, and changes implemented in production. [3]
- A source code repository with an audit trail of material changes. [3]
The monitoring standards required continuous real-time monitoring by knowledgeable and qualified staff while systems were trading. [3] Monitoring staff would need the ability and authority to disengage a system and cancel resting orders. [3] Automated alerts would fire when order behavior breached design parameters or connectivity or data feeds were lost. [3]
What the Commission thought the lifecycle would cost
The proposal's cost discussion made a notable assumption. [3] It said the Commission had calculated costs for certain risk mitigation procedures but determined they generally would not be imposed. [3] Among its reasons, it said they related to procedures or controls already widely used in the industry. [3] It gave the testing, monitoring and supervision procedures as an example. [3]
It said the Commission believed the risk controls required of trading firms and clearing firms were already widely used by market participants. [3] It expected upgrades to impose one-time costs but generally not increased annual costs. [3] In our reading, that assumption cut both ways. If the controls were already in use, the rule's benefit lay mainly in standardization and inspection. Opponents later argued the benefits were not commensurate with the costs. [5]
Periodic review
The proposal closed each set of standards with a review duty. [3] Each AT Person would periodically review its compliance with the risk control requirements to determine whether it had effectively implemented sufficient measures reasonably designed to prevent an Algorithmic Trading Event. [3] It would take prompt action to remedy deficiencies it found. [3] The development and testing policies would also be reviewed periodically, with deficiencies documented and remedied. [3]
In our reading, those review duties are where a firm's internal audit or second line would have entered. They imply a written review, a list of findings and a record of remediation. Those artifacts would feed the annual compliance report to the exchange. [2]
Testing on each venue
The proposal required testing both internally and on each exchange where algorithmic trading would occur. [3] Exchanges, in turn, would provide test environments where AT Persons could test compliance with the risk control and development requirements. [2]
For a firm trading on several venues, that meant a test plan per venue. In our reading, the requirement recognized that an algorithm's behavior depends on the venue's matching rules, message limits and data feeds. A test that passes in a firm's simulator can fail against a venue's actual behavior.
Why the Commission pointed at the lifecycle
The proposal explained its concern with examples of failures. [3] It cited the joint CFTC and SEC report on the May 2010 Flash Crash, which found that an automated execution algorithm did not take price or time into account. [3] It cited a 2012 event at Knight Capital Group, which it said involved a coding error and incorrect deployment of new code in an automated equity router. [3]
The proposal said those errors led to losses of more than $460 million for the firm. [3] In our reading, the Knight example explains the emphasis on deployment controls, since the failure the proposal described was in how code reached production.
Events outside the United States
The proposal also described disruptions on foreign markets. [3] It said that in February 2015 the opening of futures and options trading on Eurex in Germany was delayed for five minutes because a participant's system was transmitting duplicate orders. [3] It said that in February 2014 trading in three-year Korean treasury bonds was halted for almost two hours at the Korea Exchange after a system malfunction from an improper order. [3] It said that on October 26, 2011 the Bombay Stock Exchange canceled all derivatives trading after a member firm's flawed algorithm. [3]
In our reading, the common thread in these examples is that one participant's system affected everyone on the venue. That is the risk the exchange layer addresses directly, and the one the 2021 principles kept in view. [6]
The people requirement
The fact sheet said the standards covered the designation and training of algorithmic trading staff. [2] In our reading, that tied the controls to people: someone qualified had to be watching, with authority to stop the system.
For a firm, this is a staffing commitment for every hour its algorithms trade. A firm trading around the clock on electronic platforms would need monitoring coverage around the clock, or a policy to disengage algorithms when no one is watching. The proposal left the design to the firm. [3]
What the lifecycle asks of a bank
In our reading, the lifecycle standards resemble the general IT change controls a bank's technology risk function already knows: segregated environments, pre-production testing, change records and monitoring. The proposal's novelty was applying them specifically to trading algorithms and making them inspectable by exchanges and the Commission.
The cost lay in documentation. A firm whose algorithms were developed by traders with informal testing would need to rebuild its process. A firm with mature software controls would mainly need to show them in the form the proposal asked for.
Registered futures associations as a fourth layer
The proposal added a self-regulatory layer. [1] [2] All AT Persons would have to become members of at least one registered futures association. [2] Those associations would adopt membership rules relevant to algorithmic trading, as they deemed appropriate, for each category of member. [2]
The fact sheet said this would allow associations to supplement elements of Regulation AT as markets and trading technologies evolve. [2] The press release said the provisions together would help keep the regulation current. [1] In our reading, this was a design for change: detailed rules could be updated by an association without a new Commission rulemaking.
The source code record was where the proposal broke
The 2015 proposal treated the source code repository as a record. [3] Each AT Person would keep it and make it available for inspection under 17 CFR 1.31, the Commission's general recordkeeping rule. [3]
What the record contained
The repository would manage source code access, persistence, copies of all code used in production, and changes to that code. [3] Its audit trail would show, for each material change, who made it, when, and the coding purpose of the change. [3] In our reading, this is an ordinary engineering record in any well-run development shop.
Why access became the issue
The objection was not to keeping source code. It was to who could see it, and how. The 2016 supplemental proposal acknowledged the proprietary value of algorithmic trading source code and revised the provisions extensively. [4] It said source code and related records would still have to be preserved, but would no longer be subject to the general recordkeeping provisions. [4]
Under the supplemental proposal, the Commission would have access to source code only via a subpoena or a special call approved by the Commission itself, not by staff. [4] Access would be subject to policies and procedures to protect confidentiality. [4] The supplemental proposal also added a definition of "Algorithmic Trading Source Code," a term the original proposal did not use. [4]
The concession did not save the proposal. [5] The 2020 withdrawal recorded that commenters on the original proposal had opposed source code preservation and access without a subpoena. [5] It said the Commission would not pursue proposals that would compel market participants to divulge their source code and other intellectual property absent a subpoena. [5]
How a firm would have organized the repository
In our reading, the repository requirement described a familiar tool: version control, which records who changed code, when, and with what message. The proposal's audit trail asked for the coding purpose of each material change. [3] For a development team, that means change messages that explain intent.
The harder parts were scope and retention. The repository had to hold copies of all code used in production. [3] For a firm whose algorithms depend on shared libraries, vendor components and configuration files, deciding what counted as the algorithm's code would have taken judgment. The supplemental proposal's new definition of algorithmic trading source code was, in part, an answer to that question. [4]
What the dispute teaches about records
For any regulated firm, the source code fight raises a general question. A record a firm keeps for its own control purposes can become a regulator's evidence once a rule names it. In our reading, the right response is to decide in advance how such a record is classified, who in the firm can release it, and under what legal process.
That decision is a governance matter. Legal, technology and compliance should agree on it before a regulator asks, whatever the rule says at the time.
What survived was the exchange's control
The CFTC withdrew Regulation AT effective July 15, 2020. [5] The withdrawal said the Commission had proposed it based on certain assumptions about the relative risk of automated or algorithmic trading compared with other forms of electronic trading. [5]
The three rejected approaches
The withdrawal listed three provisions and rejected each as a means of addressing the perceived risk underlying the proposals. [5] The first required certain market participants to register based on their trading functionality, strategies or market access methods. [5] The second compelled registrants, including newly registered participants, to produce source code without a subpoena. [5] The third applied prescriptive requirements for the types of risk controls that exchanges, futures commission merchants and others would implement. [5]
It said the Commission had determined not to proceed with detailed, prescriptive requirements such as those in the proposals. [5] It said it would not pursue proposals that required additional classes of participants to register. [5]
The replacement
The Commission adopted the Electronic Trading Risk Principles as a final rule published January 11, 2021. [6] It set three principles for exchanges. [6] An exchange must adopt and implement rules governing market participants to prevent, detect and mitigate market disruptions or system anomalies associated with electronic trading. [6] [7] It must subject all electronic orders to exchange-based pre-trade risk controls for the same purpose. [6] [7] It must promptly notify Commission staff of any significant market disruptions on its platforms and provide timely information on causes and remediation. [6] [7]
The rule took effect January 11, 2021, with full compliance due no later than July 12, 2021. [6] Its acceptable practices said an exchange could comply by adopting rules and risk controls reasonably designed to prevent, detect and mitigate such disruptions. [6]
What changed in the architecture
Set side by side, the two designs differ in where the duty sits. Regulation AT placed duties on three layers and reached the firm directly. [2] The 2021 rule placed duties on the exchange, which the CFTC already regulates, and left the exchange to set rules for participants. [6] In our reading, the firm's obligations now flow mainly through exchange rules rather than CFTC regulation.
The scope also changed. Regulation AT was built around algorithmic trading. [2] The 2021 principles speak of electronic trading and all electronic orders. [6] [7] In our reading, that ended the definitional problem of separating algorithmic from manual electronic orders.
What exchanges must show now
The 2021 rule's acceptable practices use a reasonableness standard. [6] They provide that an exchange can comply by adopting and implementing rules and risk controls reasonably designed to prevent, detect and mitigate market disruptions and system anomalies. [6] [7] The rule lists no particular control. [7]
The third principle concerns notification. [6] [7] An exchange must promptly notify Commission staff of any significant market disruption on its platform, with timely information on causes and remediation. [7] The final rule recorded commenters' disagreement over whether "significant" was clear enough to enforce, and other commenters' support for the term, one of them saying it gave exchanges discretion. The Commission said the standard was workable in practice. [6]
In our reading, the notification principle gives the Commission an event feed without a reporting form. Exchanges apply the significance threshold in the first instance. For a trading firm, a disruption it causes becomes a matter the exchange may report to the Commission, which raises the stakes of its own controls.
The strongest case for the original design
A fair objection is that the withdrawal gave up direct oversight of trading firms, the place where the failures the proposal cited began. [3] The Knight and Flash Crash examples involved firm-level code and algorithms. [3] Exchange controls can stop a runaway order stream; they cannot inspect the code that produced it.
The objection has force. The answer the Commission gave in 2020 was that the approaches needed to reach the firm, registration and source code access, were not justified. [5] Whether exchange controls suffice is a judgment about risk, and the sources we opened do not measure outcomes either way. We make no claim that disruptions have or have not occurred since.
A second objection
Another objection is that principles-based rules for exchanges leave each venue to set its own standard, so a firm faces different controls on different venues. The 2021 rule's text supports the premise: it requires exchanges to adopt rules and pre-trade risk controls for the purpose, and its acceptable practices speak of controls reasonably designed for it, without listing them. [6] [7]
In our reading, the variation is the price of flexibility, and firms bear it. A firm trading on several venues has to learn each venue's controls and keep its own limits inside the tightest. That is the same reconciliation work the three-layer design would have required, now driven by exchange rules instead of a CFTC list. [2] [6]
What the withdrawal did not reject
The withdrawal named what it rejected. [5] It did not reject pre-trade risk controls as such, or the idea that exchanges should apply them. [5] [6] Its target was the reach to new registrants, compelled source code and prescriptive control lists. [5] In our reading, that left the engineering discipline of the 2015 proposal as an open recommendation for firms to adopt on their own judgment, rather than a requirement the Commission had disowned.
What we would tell a trading or clearing firm today
Build the lifecycle controls Regulation AT described, because they are sound engineering, not because a rule requires them. [3] Know the exchange-based pre-trade controls on every venue you trade, and test how your systems react when an order is rejected. [6] Decide how source code is classified and who may release it, before anyone asks.
For a clearing firm, the layering question has not gone away. Exchange rules may require clearing members to use exchange-provided controls for customers' orders. In our reading, the clearing firm should know those controls' parameters for each customer, since its credit exposure is what they protect.
With hindsight
The proposal approved on November 24, 2015 was published in the Federal Register on December 17, 2015. [3] The Commission issued a supplemental proposal on November 25, 2016, with comments due on or before January 24, 2017. [4] [5] It proposed modifications to the risk control framework, trading firm registration criteria, reporting requirements and source code provisions, and compliance options for firms using third-party algorithmic trading systems. [5]
The withdrawal notice recorded the comment history. [5] It said that on June 10, 2016, Commission staff held a public roundtable on elements of the proposal, and the Commission reopened the comment period and received 19 additional comment letters. [5] It said all of them also expressed concern with Regulation AT. [5] It said the Commission received 27 comment letters on the supplemental proposal. [5]
On July 15, 2020, the Commission withdrew both proposals as of that date. [5] It specifically rejected new registration requirements based on trading functionality or market access, compelled source code production without a subpoena, and prescriptive risk control requirements across exchanges, clearing firms and others. [5]
The Electronic Trading Risk Principles followed: the Commission issued the final rule on December 10, 2020, and it was published January 11, 2021, effective that day, with full compliance required by July 12, 2021. [6]
The current text of 17 CFR 38.251, as checked on 2026-11-16, carries the three principles in paragraphs (e), (f) and (g). [7] It requires exchanges to adopt rules on market participants, to subject all electronic orders to exchange-based pre-trade risk controls, and to notify Commission staff promptly of significant disruptions. [7] The section's source note cites the amendment at 86 FR 2071, January 11, 2021. [7]
The withdrawal also recorded the main lines of opposition to the original proposal: to its risk control framework, to identifying and registering AT Persons, to source code access without a subpoena, and to prescriptive one-size-fits-all rules. [5]
What held up from November 2015 was the idea of exchange-based pre-trade controls on electronic orders. [2] [7] Regulation AT would have required them at exchanges, among other layers. [2] The 2021 rule requires them for all electronic orders. [6] [7]
What did not hold up was the reach to the firm. [5] The registration of proprietary traders, the source code access and the prescriptive control lists for every layer were each rejected by name. [5] The development and testing standards that would have applied to trading firms do not appear in the current section 38.251. [3] [7]
The 2021 rule's own background explains its framing. [6] It described the principles as applying to market disruptions and system anomalies associated with the entry of electronic orders and messages into exchanges' platforms. [6] It said such disruptions originating at a market participant could limit other participants' ability to trade, engage in price discovery or manage risk. [6] It recounted that the proposal had described electronic trading as including all trading and order messages submitted electronically to an exchange's platform, both automated and manual order entry. [6]
The proposal's engineering standards remain a useful description of how a trading firm can control its own algorithms. [3] They are no longer a pending CFTC requirement. [5] For firms, the obligations that apply now come through exchange rules adopted under the 2021 principles. [6] [7]
Spotted an error? Tell us and we will check it and correct the article where needed.