OFR’s repo rule put LIBOR’s successor on a next-morning data feed

In February 2019 the OFR made a clearinghouse’s daily repo file a mandatory federal collection that feeds SOFR. For banks leaving LIBOR, the rate became a dependency with a filing deadline, a revision window and a contingency method, and systems needed designing around all three.

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

In February 2019 the OFR made a clearinghouse’s daily repo file a mandatory federal collection that feeds SOFR. For banks leaving LIBOR, the rate became a dependency with a filing deadline, a revision window and a contingency method, and systems needed designing around all three.

  1. The week On February 20, 2019, the Federal Register published a final rule from the U.S. Department of the Treasury’s Office of Financial Research (OFR) titled “Ongoing Data Collection of Centrally Cleared Transactions in the U.S. Repurchase Agreement Market” (84 FR 4975, codified in 12 CFR Part 1610).
  2. A rate published after the day it measures The 2019 rule sets the deadline at the source: data on repo transactions must reach the New York Fed no later than 6:00 a.m. Eastern time on the business day following the transaction.
  3. A same-day window in which the rate can change The second property is revision.
  4. A supply chain the bank does not run The third property is supply.
  5. Identifiers inside the calculation The fourth property is identity.
  6. Fallback language as data in the system of record The fifth property is how contracts consume the rate.
  7. The segment the 2019 rule left out The sixth property is coverage.
  8. A design review for systems that consume SOFR The six properties can be turned into a review that a bank’s architecture function can run on any system that reads the rate.
  9. With hindsight The collection became the rate’s production source.

The week

On February 20, 2019, the Federal Register published a final rule from the U.S. Department of the Treasury’s Office of Financial Research (OFR) titled “Ongoing Data Collection of Centrally Cleared Transactions in the U.S. Repurchase Agreement Market” (84 FR 4975, codified in 12 CFR Part 1610). The rule set an effective date of April 22, 2019. It requires daily reporting to the OFR by covered central counterparties, or CCPs, on the repurchase agreement (repo) transactions they clear.

The rule states two purposes. The first is to help the Financial Stability Oversight Council, its member agencies and the OFR identify and monitor risks to financial stability in the repo market. The second is to support the calculation of certain reference rates. The rule names them: the Secured Overnight Financing Rate (SOFR), which it describes as the Alternative Reference Rates Committee’s preferred alternative to U.S. dollar LIBOR, and the Broad General Collateral Rate (BGCR). The preamble explains that SOFR relies on data on Treasury-backed repos from three segments of the market. The Federal Reserve Board collected tri-party data through its supervisory authority over the clearing banks. Some data on the two centrally cleared segments were available to the Federal Reserve Bank of New York through a voluntary agreement with DTCC Solutions LLC, an affiliate of the clearing corporation that runs them. A mandatory collection, the OFR wrote, would increase confidence that the rate’s inputs would continue to be available, especially if new CCPs entered the market.

The scope is set by activity. A CCP is covered if its average daily total open commitments in repos, across all its services and over all business days of the prior calendar quarter, are at least $50 billion. The OFR said it believed only one reporter was then covered: the Fixed Income Clearing Corporation (FICC), a subsidiary of the Depository Trust & Clearing Corporation. It expected reporting at first from two FICC services, the General Collateral Finance (GCF) Repo Service and the Delivery-Versus-Payment (DVP) Service. The rule says the collection covers centrally cleared repo making up approximately one-quarter of all U.S. repo market transactions.

The mechanics are specific. The Federal Reserve Board acts as the OFR’s collection agent, and covered reporters submit directly to the New York Fed, which passes the data to the OFR. To produce SOFR and BGCR, the rule says, transaction data must reach the New York Fed no later than 6:00 a.m. Eastern time on the business day after the transaction, through a process allowing for “the secure, automated transmission of files.” The regulation lists the required fields in three tables: general collateral trades, the collateral delivered against net general collateral exposures, and specific-security trades. They include submission and match timestamps, the rate on an actual/360-day basis, principal, start and end dates, and the legal entity identifier (LEI), legal name and internal identifier of each cash provider, securities provider, clearing member and broker.

Implementation came in three phases after the effective date: specific-security data within 180 days, general collateral data within 240 days, and LEIs of other parties within 420 days. The LEI phase applied if the covered reporter could effect any rulemaking through the SEC needed to require market participants to obtain LEIs; otherwise it would report legal names or internal identifiers for participants without one. A commenter proposed the ISO 20022 messaging standard; the OFR declined to reference it, saying the collection is a single readout of all transactions within a period, not a transaction flow. The OFR estimated an annual burden of 1,512 hours per covered reporter.

The OFR’s February 2019 rule reads at first like a narrow statistical regulation. It was expected to bind one clearing corporation, sets out three tables of fields and explains how files reach the New York Fed. Bank technology teams could have filed it under someone else’s problem. We think it deserves a second look, because it fixed in regulation the supply chain behind the rate that U.S. dollar contracts were being moved onto.

LIBOR was a panel rate. According to the Federal Reserve Board, it was calculated from submissions by a panel of large banks and published for tenors from overnight to 12 months. SOFR, as the 2019 rule makes plain, is a calculation over transaction files. One of those files comes from a clearinghouse under a federal reporting rule with a 6:00 a.m. deadline. The New York Fed then publishes the rate, revises it on the same day if errors exceed a threshold, and falls back to an adjustment method if a segment’s data do not arrive. Each of those steps is a property of an upstream system that a bank does not run but whose output its loan, deposit, hedging and finance systems consume.

Our argument is that this rule turned a benchmark into a data dependency, and that the architecture of rate-dependent systems should be designed around the dependency’s published behavior, not around assumptions carried over from a panel rate. The analysis falls into six parts, each a property of the dependency that a system design has to answer for:

  • Timing: a rate published after the day it measures.
  • Revision: a published window in which the rate can change.
  • Supply: a chain of reporters and agents that the bank does not control, with a stated contingency method.
  • Identity: legal entity identifiers as inputs to the calculation itself.
  • Contract terms: fallback language that has to live in the system as data.
  • Coverage: the segment of the market the 2019 rule left out, and what that meant for the next collection.

The first four parts concern the rate’s production, the fifth concerns how contracts consume it, and the sixth concerns where the reporting obligation would go next. A short closing section turns the six into a design review.

A rate published after the day it measures

The 2019 rule sets the deadline at the source: data on repo transactions must reach the New York Fed no later than 6:00 a.m. Eastern time on the business day following the transaction. The New York Fed’s SOFR page completes the sequence. Each business day it publishes SOFR at approximately 8:00 a.m. Eastern time. Its methodology page says that SOFR is published one business day after its value date, the day the overnight trades were negotiated.

For a system designer, that sentence carries the whole timing problem. On any business day, the value the system can fetch describes the previous business day. A rate for an interest period that ends today cannot be complete until tomorrow morning, after the clearinghouse file has arrived and the calculation has run.

The New York Fed added tools that change where the timing burden lands. It publishes compounded SOFR averages over 30, 90 and 180 calendar days and a SOFR Index from which a compounded rate over a custom period can be calculated. Its methodology states that, unlike SOFR itself, the averages and index are published on their value date, shortly after SOFR at approximately 8:00 a.m. Eastern time. A system that computes interest from the index needs two index values and a day count. A system that compounds daily SOFR observations itself needs every observation, a holiday calendar and the convention the contract specifies. The methodology adds a detail that matters for reconciliation: the averages and index employ daily compounding on each business day, and because the overnight trades behind SOFR mature on the next business day, the index and averages always carry a value date one business day later than the value date of the last SOFR observation they include. Two systems that compute the same period, one from the index and one from daily observations, will agree only if both apply the same dating rule. A reconciliation that flags small differences between them may be finding a dating mismatch rather than an error.

Neither choice is a technical detail. It determines when a payment amount is known, and so what a borrower statement, a payment instruction or a general ledger accrual can show and when. A system that assumes a rate is known at the start of a period has to be told where in the contract the observation period begins and ends, and what happens on days with no publication.

The New York Fed also sets out its publication calendar. Its methodology says the Treasury repo reference rates are typically published every business day, including days on which SIFMA recommends an early close for secondary trading of government securities, but not on or for days on which SIFMA recommends a full closure. It adds that the New York Fed may, with advance notice, choose not to publish on a business day that participants in the Treasury repo market broadly expect to treat as a holiday. A rate engine therefore needs a calendar that it can update from a notice, not one fixed at deployment.

The design checks for this part are concrete. A bank should be able to list which systems read SOFR or its averages and at what time of day their jobs run. It should be able to show that no job runs before the morning publication and uses the prior value without saying so. It should know whether the loan system, the hedge accounting tool and the treasury system share one holiday calendar or keep their own. And where a contract compounds in arrears, the system should hold the contract’s observation convention and record which observations it used for each period. A bank that can show all of this from configuration and logs, rather than from memory, has the timing dependency under control.

A same-day window in which the rate can change

The second property is revision. The New York Fed’s methodology says that if, after the morning publication, errors are discovered in the calculation process or in the transaction data provided by BNY or the OFR, or if data that were unavailable in time for the morning publication later become available, the affected rate may be revised at approximately 2:30 p.m. Eastern time. Revisions occur only if the change in the rate exceeds one basis point and only on the same day as the initial publication. A footnote marks any revised rate. The SOFR Averages and Index follow the same pattern and are revised on a same-day basis only, at approximately 2:30 p.m., if that day’s SOFR is being revised or there is an error in their own calculation. The page says the threshold will be reviewed periodically and may change with market conditions.

That policy is a gift to system designers, because it is bounded. A rate can change once, on one day, at a stated time, above a stated threshold. A system can be built to respect that window.

There are three broad choices. The first is to wait: schedule rate-dependent processing after the revision window closes, so that the value consumed is final. That is clean but moves work later in the day, which may collide with payment cutoffs or customer-facing commitments. The second is to consume early and reconcile: take the morning value, then compare it with the value after the revision window and re-run anything affected if it changed. That keeps processing early but requires every downstream calculation to be repeatable from stored inputs. The third is to consume early and accept the morning value, with a documented decision that a change of a basis point or more on the same day will be handled as an exception. Each choice is defensible. Making none of them is not, because the system will then behave in whichever way its job scheduler happens to dictate.

The second choice deserves a note on architecture. A recalculation is safe only if the system stores the rate value it used, the time it fetched it and the source, and if the calculation is idempotent: running it again with the corrected rate produces the corrected result without double-posting. Systems that write an accrual and forget the input cannot do this. Systems that keep an append-only record of rate observations, keyed by value date and publication time, can.

Revision also matters for evidence. If a customer disputes an interest charge, or an auditor tests an accrual, the bank needs to show which published value it used and why. The New York Fed’s footnote marks a revision on its side. The bank’s own records need to show whether a revision reached its systems, and when.

The methodology describes one more layer that data teams should know about. For SOFR, BGCR and the tri-party rate, the New York Fed publishes updated summary statistics on a lagged basis shortly after the end of each quarter, and says these may differ from the originally published data if errors are found in the data from BNY or the OFR or in the calculation. The rates themselves are revised only on the same day as their initial publication. A bank that uses the summary statistics, for example the percentiles or volumes in a market risk report, should therefore keep the vintage of each figure it uses. A dataset that overwrites the earlier statistics with the later ones will produce a report that cannot be reproduced.

A supply chain the bank does not run

The third property is supply. The 2019 rule describes a chain with named links. A covered CCP gathers trade data during processing. It sends files directly to the New York Fed, acting for the Federal Reserve Board as the OFR’s collection agent. The New York Fed passes the data to the OFR and uses it, under data sharing the Council authorized, to produce SOFR and BGCR. The New York Fed’s SOFR page describes the current inputs: tri-party repo data collected from the Bank of New York Mellon and from the OFR, and GCF Repo and DVP data obtained from the OFR.

The rule is explicit about why the OFR wanted a mandatory collection. Before it, some data on the cleared segments reached the New York Fed through a voluntary agreement with DTCC Solutions. A mandatory collection, the OFR wrote, would increase confidence that the rate’s inputs would continue to be available. It also wrote that some fields in the collection were not received under the voluntary agreement but would help ensure the continued quality of the rates, and that the identity of counterparties matters because it lets the calculation agent identify and, as appropriate, exclude transactions that may not be representative of market activity, such as certain affiliate transactions.

For a bank, this chain is a third party it has no contract with. There is no service agreement to negotiate and no right to audit the clearinghouse’s reporting system. What there is, instead, is a public record of how the chain behaves when it fails. The New York Fed’s methodology describes its data contingency for the Treasury repo reference rates. If data for a market segment are unavailable, it uses that segment’s latest available data, with each transaction’s rate adjusted by the change in market rates since then. It measures that change from a daily survey of primary dealers’ repo borrowing, in which dealers report volumes and volume-weighted average rates by segment. Segments with current data are used unadjusted. If data other than those provided by BNY or the OFR are used, the change in source is noted when the rate is published.

That contingency means a bank’s rate feed will keep arriving even if a segment’s data do not, but the value will rest partly on estimated inputs, with a note saying so. Whether that matters depends on the bank’s use. Where a loan contract defines its rate as the published SOFR, the published value is the value, whatever data sat behind it. For a model that uses SOFR as an input to valuation or to an internal transfer price, a flagged day might deserve a different treatment. Either way, the system has to be able to see the note. A feed that strips footnotes and stores only a number cannot.

The supply chain also shapes resilience planning. A bank’s business continuity plan can list the internal systems that compute interest and the vendors that deliver market data. It should also list the public sources those vendors repackage, with the published behaviors of each: the publication time, the revision window, the contingency method. Where a vendor supplies SOFR, the bank should know whether the vendor passes through revisions and footnotes, and how quickly. The 2019 rule makes the upstream chain unusually transparent. The gap that remains is between the public source and the bank’s own systems.

The rule’s implementation schedule is a small case study in how to move a production dependency from one feed to another. The one expected covered reporter cited operational complexities in the scale of the data-field builds and the testing they needed, and asked for a longer, phased timeframe. The OFR adopted three: specific-security trade data first, general collateral data second, and the LEIs of other parties last. It explained that the phases would not create a gap in the data needed to calculate SOFR, because the SOFR inputs were a subset of the new collection and the New York Fed was already receiving them under the voluntary agreement. The old supply stayed in place while the new one was built and tested. A bank should look for the same pattern when it changes its own rate source, whether moving from one market data vendor to another or from vendor-supplied values to the administrator’s published data: run both, compare, and switch only when the comparison is clean.

There is a further point about concentration. The rule’s threshold design meant that, when it was adopted, the OFR believed one CCP was covered. The rule also provides that any CCP that later meets the $50 billion threshold becomes a covered reporter, with reporting starting on a schedule the rule sets, and the OFR wrote that this would ensure rate production includes new comparable transactions as repo markets evolve. For a bank, that means the composition of the rate’s inputs can change over time without any change to the rate’s name. A rate dependency is a dependency on a methodology, and methodology pages have change histories worth watching.

Identifiers inside the calculation

The fourth property is identity. The 2019 rule requires the LEI, legal name and internal identifier of each party to a cleared repo: the cash provider and securities provider, the clearing members through which each accessed the service, and any broker. It requires the LEI of the covered reporter itself. The rule says LEIs reported must satisfy the standards implemented by the Global LEI Foundation.

The OFR’s reasons were practical. It wrote that counterparty identity is important for the rate calculation, because it allows the calculation agent to identify and, as appropriate, exclude transactions that may not represent market activity. It also wrote that mandatory LEI reporting would benefit firms and regulators. And it accepted the covered reporter’s request for a longer runway: LEIs of parties other than the reporter would be reported within 420 days of the effective date, and only if the reporter could effect any SEC rulemaking needed to require participants to obtain LEIs. Where it could not, the reporter would report legal names or internal identifiers for participants without an LEI.

For a bank that clears repo, directly or through a sponsoring member, the rule set out a path by which its LEI would travel with its trades into a federal collection, if the clearinghouse could put an LEI requirement in place through an SEC rulemaking; otherwise its legal name or an internal identifier would. Reference data that is stale, or held differently in the trading system and the legal entity master, becomes visible to a regulator at the transaction level.

For a bank that only consumes SOFR, the point is narrower but still real. A rate whose inputs are filtered by counterparty identity depends on identity data being right. Banks have little influence over that, but they have full control over the same discipline in their own data. The 2019 rule is a clear example of a regulator building a calculation on top of a global identifier.

The rule also tells participants what happens to their trade data once it is reported. The OFR wrote that it would make the data available to the Council and its member agencies as necessary to support their regulatory responsibilities, subject to the confidentiality and security requirements of applicable laws, including the Dodd-Frank Act. It noted that, under that Act, submitting nonpublic data to the OFR does not waive any privilege that attaches to it. It said aggregate or summary data might be published to increase market transparency, to the extent that intellectual property rights are not violated, confidential business information is protected and publication poses no significant threat to the financial system. For a bank’s data governance function, this is a map of where its transaction-level data travels once it clears a trade, and it belongs in the bank’s own record of processing alongside the flows the bank controls. We would expect the same pattern in other collections, and bank data teams can treat LEI quality as infrastructure rather than as a compliance form.

The OFR’s response to the ISO 20022 comment adds a second architectural lesson. A standards advisory group proposed its own messaging standard. The OFR declined to reference it, explaining that ISO 20022 is a transaction messaging standard while the collection is based on a single readout of all transactions within a period. That distinction, between a feed of events and a periodic snapshot, is worth carrying into a bank’s own design. Rate observations, regulatory files and reconciliations are snapshots with an observation date. Payments and trades are events. A data platform that treats both alike will struggle to answer the question an examiner or auditor asks: what did you know, as of when?

Fallback language as data in the system of record

The fifth property is how contracts consume the rate. The 2019 rule did not touch contracts. It secured the inputs to the rate that the Alternative Reference Rates Committee had chosen as LIBOR’s replacement. The contracts themselves were left to the parties, and later to Congress and the Federal Reserve Board.

The Board’s final rule implementing the Adjustable Interest Rate (LIBOR) Act, published in January 2023 as Regulation ZZ, shows what the system of record eventually had to hold. The rule establishes benchmark replacements for contracts governed by U.S. law that reference the overnight and one-, three-, six- and 12-month tenors of U.S. dollar LIBOR and that do not have terms providing for a clearly defined and practicable replacement after the first London banking day after June 30, 2023. For derivatives, it points to the rate identified as the “Fallback Rate (SOFR)” in the ISDA protocol. For other contracts it uses SOFR in place of overnight LIBOR and the matching tenor of CME Term SOFR in place of the others, with separate provisions for consumer loans and for contracts of entities regulated by the Federal Housing Finance Agency, some of which use 30-day Average SOFR. It restates the statutory tenor spread adjustments: 0.644 basis points for overnight LIBOR, 11.448 for one-month, 26.161 for three-month, 42.826 for six-month and 71.513 for 12-month.

Each of those items is a field. A loan system has to know which tenor a contract referenced, whether its fallback language was clearly defined and practicable or was replaced by operation of law, which replacement rate applies, which spread adjustment applies, and from which date. For consumer loans, the Act and the rule add a one-year transition in which the spread moves linearly, each business day, from the difference between the replacement and LIBOR on the day before the replacement date to the statutory spread. A system that applies that schedule needs both its start values and its end date. The rule also says that rates published by Refinitiv Limited as “USD IBOR Cash Fallbacks” for “Consumer” products are deemed equal to the consumer replacement rates it sets out. For a system designer, that clause matters because it names a published feed that the regulation treats as equivalent to its own formula, which gives a bank two ways to obtain the value and a basis for checking one against the other.

In 2019 those fields did not exist in a statute. What did exist, and what the OFR rule made more durable, was the rate itself. A bank that in 2019 began to model fallback terms as structured data, rather than as clauses in scanned documents, put itself in a position to apply whatever the final answer turned out to be. A bank that left fallback terms in documents faced a review of every affected contract when the answer arrived.

This part also has an operational edge. Fallback terms change the rate a contract pays, so a misclassified contract produces a wrong payment. The rule’s list of contract categories, together with its separate treatment of consumer loans, made classification a control. A design that records the classification, the reason and the source document for each contract, and that can be re-run if the classification is challenged, makes that control testable.

The segment the 2019 rule left out

The sixth property is coverage. The 2019 rule says the cleared segments it reached made up approximately one-quarter of U.S. repo market transactions, and that the New York Fed collected data on approximately another quarter under the Board’s supervisory authority. The OFR called the rule an important step toward fully addressing the Council’s recommendation for a permanent repo collection, not the whole of it. The OFR’s page on the collection describes it as the first time the OFR went directly to industry to collect financial market information on an ongoing basis, and says the collection was expected to begin in October 2019.

The missing piece was bilateral repo that is not centrally cleared. The 2019 rule did not reach it, because the rule collected from CCPs, which, the OFR wrote, “already obtain most or all of the requested data during trade processing.” Bilateral uncleared trades have no such collection point. Reaching them means collecting from the firms themselves.

That design difference matters to bank architecture, because it decides who carries the reporting build. Under the 2019 rule, a single clearinghouse built the files, and its members could be required to supply LEIs. Under a collection from firms, each covered firm builds its own. The hindsight section shows that this is what happened, and what trade associations for some of those firms said about it.

The 2019 rule’s own burden estimate is a useful reference point. The OFR estimated an annual burden of 1,512 hours for a covered reporter, based on approximately three hours of reporting time for each of the general collateral and specific-security submissions on each business day, and put the total estimated annual cost at $149,688. Those figures describe recurring reporting by an organization that already held the data in its clearing systems. They say nothing about the cost of building a capture and reporting process in an organization whose trade records were not built for transaction-level reporting. A bank estimating its own reporting obligations should keep those two kinds of cost apart: the daily run cost, which a regulator can estimate from file sizes and hours, and the build cost, which depends on the state of the bank’s own trade and reference data.

Coverage has a second edge for rate users. SOFR is built from Treasury repo data in three segments: tri-party, GCF and DVP. Uncleared bilateral repo is not among them. That is a methodological choice, documented by the administrator, and it is part of what a bank relies on when it relies on the rate. When a bank’s treasury function talks about funding markets, and its systems talk about SOFR, the two are not describing the same population of trades. A risk report that conflates them can mislead.

The administrator’s own definition shows how much filtering and selection lies between the market and the number. The New York Fed describes SOFR as a broad measure of the cost of borrowing cash overnight collateralized by Treasury securities. It includes all trades in the Broad General Collateral Rate plus bilateral Treasury repo cleared through FICC’s DVP service, filtered to remove a portion of transactions considered “specials”, which the page explains are repos for specific-issue collateral that take place at cash-lending rates below those for general collateral repos. The rate is calculated as a volume-weighted median of the transaction-level data. Each of those choices, the segments, the specials filter and the median, is a documented property of the dependency. A bank does not need to agree with them. It needs to know them, record which version of the methodology applied when it relied on the rate, and notice when they change.

The same discipline applies to the bank’s own internal rates. Where a treasury function builds a transfer pricing curve or a discounting curve from SOFR and other inputs, the curve is a derived dataset with its own methodology, inputs and revision behavior. The 2019 rule and the administrator’s pages are a model of how to document such a thing: what goes in, from whom, by when, what is excluded and why, what happens when an input is missing, and when the output can change. An internal curve documented to the same standard is easier to defend to a model risk reviewer than one whose method lives in a spreadsheet.

A design review for systems that consume SOFR

The six properties can be turned into a review that a bank’s architecture function can run on any system that reads the rate. We would frame it around evidence rather than intention.

  1. Inventory: list every system that reads SOFR, its averages, its index or a term rate built on it, with the source of each feed and the time each job reads it.
  2. Timing: for each system, record whether it computes from daily observations, averages or the index, and show where the observation period, lookback and holiday calendar are configured.
  3. Revision: document which of the three revision choices applies, and test that a same-day revision of more than one basis point is either waited for, reconciled or handled as an exception.
  4. Contingency: confirm that the feed carries the publisher’s footnotes and that a day published from contingency data can be identified afterward.
  5. Identity: reconcile the LEIs held in trading, clearing and legal entity systems, and record the owner of each.
  6. Contract terms: show that fallback classification, replacement rate, spread adjustment and effective date are held as fields with a source document and an audit trail.

Each step produces something an examiner, auditor or internal audit team can test. The inventory is a list. The timing and revision decisions are documented configurations with test results. The contingency check is a query that returns the days published from contingency data and shows how each system treated them. The identity reconciliation is a report with exceptions and owners. The contract terms are fields with lineage to documents. These steps need no new technology. They need the rate to be treated as an external dependency with a published contract of behavior, which is what the 2019 rule and the administrator’s methodology together provide. The bank’s side of that contract is to write down how its systems respond.

There is a staffing implication too. A rate dependency crosses teams: treasury chooses conventions, lending operations sets up loans, finance books accruals, market data teams manage feeds and architecture sets the platform. In our view, the review works only if one named owner can answer for the whole chain inside the bank, from the morning publication to the posted entry. The 2019 rule named the upstream owners. The downstream ones are for each institution to name.

Finally, the review should be repeated when the upstream record changes. The methodology page says the revision threshold may change. The rule says new CCPs can become covered. The OFR later extended collection to new segments. Each change is a public notice, and each one is an occasion to re-run the inventory. A bank that subscribes to those notices, and routes them to the system owners, turns an external change into an internal ticket, which is the point of having the inventory at all.

With hindsight

The collection became the rate’s production source. The New York Fed’s methodology page now says that before January 24, 2022, transaction data for the Treasury repo reference rates were collected under the Board’s supervisory authority and from DTCC Solutions under a commercial agreement. Its SOFR page describes the current inputs as tri-party data collected from the Bank of New York Mellon and the OFR, and GCF Repo and DVP data obtained from the OFR. The voluntary arrangement the 2019 rule described as a reason for a mandatory collection has been replaced by it.

LIBOR’s end came through law as well as markets. The Federal Reserve Board’s Regulation ZZ, published on January 26, 2023 and effective February 27, 2023, implemented the Adjustable Interest Rate (LIBOR) Act. It set SOFR-based replacements for U.S.-law contracts referencing the overnight and one-, three-, six- and 12-month tenors of U.S. dollar LIBOR that lacked a clearly defined and practicable replacement after the first London banking day after June 30, 2023. The rule notes that until December 31, 2021, LIBOR’s administrator calculated and published LIBOR for five currencies and seven tenors. For derivatives the rule adopts the ISDA protocol’s fallback rate; for other contracts it uses SOFR or CME Term SOFR, with the statutory tenor spread adjustments and a one-year transition for consumer loans. For contracts without workable fallbacks, the replacement rate and the spread became a matter of law.

The OFR then reached the segment its 2019 rule left out. On May 6, 2024, it published a final rule for daily reporting of non-centrally cleared bilateral repo, effective July 5, 2024. According to the OFR’s 2025 amendment, reporting is required from two categories of firms: brokers and dealers whose average daily outstanding commitments to borrow cash and extend guarantees in such trades are at least $10 billion, and other financial companies with over $1 billion in assets or assets under management and at least $10 billion of such commitments with counterparties that are not brokers or dealers. The compliance date for the first category occurred on December 2, 2024.

The second category asked for more time, and its reasons are an architecture brief. In an amendment published on April 2, 2025, the OFR granted a 90-day extension of the Category 2 compliance date, moving it from April 1, 2025 to June 30, 2025. It recorded that trade associations had described challenges including the automation of reporting processes for financial companies not always accustomed to transaction reporting, the development of systems that both capture and report trade information, computing reporting thresholds, and a perceived limited number of third-party service providers to submit data for them. It also recorded that potential Category 2 reporters had connected to the OFR’s Data Collection Utility and submitted test files. The OFR said it did not believe an extension of the lengthier periods requested was warranted, and that reporting by Category 1 firms had already started.

A separate rulemaking widened clearing. On December 13, 2023, the SEC adopted a requirement that covered clearing agencies for Treasury securities have policies requiring their direct participants to submit eligible secondary market transactions, including eligible repo, for clearing. In a release published March 4, 2025, it extended the compliance dates by one year, to December 31, 2026 for eligible cash market transactions and to June 30, 2027 for eligible repo. As checked on November 15, 2026, the SEC’s Treasury clearing implementation page still gives those two compliance dates. The same page says the Commission has issued exemptive relief from the clearing requirements for certain inter-affiliate and non-U.S. repo transactions. If more repo moves into clearing, more of it passes through the collection point the 2019 rule built.

What held up is the rule’s premise that a reference rate is only as dependable as its data supply. What the 2019 rule could not show is how much of the later work would fall on firms’ own systems once reporting moved from one clearinghouse to individual firms.

Spotted an error? Tell us and we will check it and correct the article where needed.

  • SOFR
  • LIBOR transition
  • Office of Financial Research
  • repo market
  • reference data
  • legal entity identifier