The Autocatalytic Sets of the Financial Markets, Two

The formal condition names a role, not an occupant

First, the termsWim Hordijk and Mike Steel study collections of chemical reactions that keep themselves going. A reaction turns some molecules, its reactants, into others, its products; the reactants are used up. A catalyst is a molecule that makes a reaction run far faster without being used up itself, the way a matchmaker brings two people together and walks away unchanged. The food set is the raw material supplied from outside. A collection of reactions that meets their condition is called a RAF, short for "reflexively autocatalytic and food-generated."

They state the condition in two halves. The first is reflexive autocatalysis: "Each reaction r ∈ R is catalyzed by at least one molecule type that is either a product of R or is present in the food set F". In plain words: every reaction in the set has a helper, and that helper is either made by the set itself or comes in with the food. The second half is that the set is food-generated: "All reactants involved in reactions in R can be created from the food set F by using a series of reactions only from R itself". In plain words: everything the reactions use up can be built, step by step, from the raw material, using only the set's own reactions. A set meeting both halves keeps itself going for as long as the food keeps arriving.

A role, not an occupantRead the first half slowly. The condition is at least one catalyst. It does not say which molecule. A reaction needs a catalyst; it does not need a particular catalyst. It works like a job description that says "a licensed electrician" and not "Bob": anyone who holds the licence fills the position. That is what makes substitution possible at all. In a financial loop, lending needs a reference rate to price loans against, not one particular rate, which is why a rate that was retired could be swapped for another one, as the first substitution below describes.

On its own, though, filling the role is not enough.

A single molecule can play more than one part. Take a set of two reactions. In the first, molecule B is the catalyst: it helps A turn into C. In the second, B is a reactant: it is used up, combining with C to make more A. Suppose A is not in the food supply, so the second reaction is the only way more A ever gets made.

Now remove B and bring in a new molecule, X, that catalyzes the first reaction just as well. The first reaction is fine; its role has a new occupant. The second reaction is not. It still needs B itself as an ingredient, and a catalyst cannot stand in for an ingredient. The second reaction stops, no more A is made, and once the existing A runs out the first reaction stops too. The vacancy was filled, and the loop still broke.

Markets have members who play more than one part in the same way. An institution can be the lender in one stretch of a loop and the borrower in another. A replacement lender fixes the first stretch and leaves the second empty. So the real test is not on the role that was vacated. It is on the whole system after the change.

a member is replaceable only if the reaction set that remains is still reflexively autocatalytic AND still food-generated → the vacated role having another occupant is necessary, and is not sufficient

That is the licence and its limit together, and it gives a two-part test for any replacement. First, does every reaction still have a helper from inside the set or from the food? Second, can everything the reactions use up still be built from the food by the set's own reactions? Only if both answers are yes, with the newcomer in place, does the loop keep closing.

Notice what the test does not ask: whether the new member is better. A replacement that is worse than the original, slower, costlier or less reliable, still passes if both answers are yes, and the loop will carry it. That is the subject of the next section.

The proposition, and the correction it needs

The proposition this page was written to examine: once a loop is established, a member can be substituted, and substitutions are admitted when they enhance the loop and rejected when they do not, because the attractor will not hold a member that makes it worse.

The first half survives. The second half does not, and the way it fails is more useful than the claim would have been.

What the condition actually requiresClosure. If the replacement fills the role, the loop keeps running, and nothing in the condition asks whether it fills it better. A substitution that degrades the loop while still closing it is admissible in exactly the sense the formalism defines, and the loop will carry it.

So there are three outcomes, not twoA replacement that closes the loop and improves it is adopted and tends to spread, because the members have reason to prefer it. A replacement that closes the loop and degrades it can still be installed, and usually has to be installed by something with authority, because nobody inside the loop wants it. A replacement that does not close the loop fails, and that failure is the only one the attractor itself performs.

Why the distinction is worth keepingEnhancement is a sufficient reason for a substitution to stick. It is not a necessary one. Collapsing the two invites the reading that whatever survived must have been an improvement, which is the same error as reading a market's recovery as proof that the market did the recovering. The second substitution below is a case where the replacement stuck and the loop it fed got measurably worse.

Substitution one: the number that meters credit

What a reference rate isWhen a bank makes a loan whose interest rate goes up and down over time, the contract does not name a fixed rate. It names a published number, a reference rate, and adds a margin: "the reference rate plus 2 percent," reset every month or quarter. Mortgages, business loans, bonds and enormous numbers of trading contracts were written this way. For decades the reference rate for dollar contracts was LIBOR, the London Interbank Offered Rate. At its peak it sat underneath contracts worth roughly $200 trillion.

The first page describes a loop of collateral and leverage: lenders lend against assets pledged as security, the borrowed money buys more assets, rising asset prices make the pledged assets worth more, and that supports more lending. Every loan in that loop is priced off a reference rate, so the rate sets how cheaply the loop can turn.

Is the rate really a catalyst?This page treats the reference rate as the loop's catalyst, and that is a proposed mapping and not an established one. The objection deserves to come first. The tempting argument is that a catalyst speeds a reaction without being used up, and a rate is not used up by the loans it prices; change it, and credit flows faster or slower while the assets themselves stay the same. The trouble is that the same description fits every tax, every fee, every capital charge a regulator sets, and every price posted on a board: none of them is used up, and all of them change how fast things move. If all of those count as catalysts, the word stops meaning anything. A reference rate is at least as well described as a setting that contracts read. This page uses the catalyst reading because the substitution teaches the same lesson either way, and it flags the reading as a reading rather than building anything on it.

Why LIBOR had to goLIBOR was not a record of actual loans. Each morning a panel of large banks reported what they estimated it would cost them to borrow from other banks without putting up any security, and the published rate was an average of those estimates. That made it easy to bend. In 2012 it came out that traders at several banks had been nudging their submissions to profit on their own positions, and the banks paid billions of dollars in fines; Barclays was first, settling for about $450 million that June. Regulators concluded the number could not be trusted. In the words of the committee set up to replace it, LIBOR "was not anchored in actual market activity, which left it vulnerable to manipulation".

The replacementThe member was replaced. The Alternative Reference Rates Committee (ARRC), convened by the Federal Reserve, "unanimously selected SOFR as its recommended alternative to USD LIBOR" in 2017, and "June 30, 2023 then marked the cessation of all USD LIBOR panel settings". SOFR, the Secured Overnight Financing Rate, published every morning by the Federal Reserve Bank of New York since April 2018, is "a broad measure of the cost of borrowing cash overnight collateralized by U.S. Treasury securities in the repurchase agreement (repo) market". In plain terms, it is the actual interest paid on more than a trillion dollars of one-night loans each day, where the borrower hands over U.S. government bonds as security, much as a pawnshop holds your watch until you repay.

So the two are not the same instrument under a different name. LIBOR was an estimate of what banks would pay to borrow from each other on trust. SOFR is a measurement of what anyone pays to borrow against the safest security there is.

What was gainedThe new number comes from real trades, not from anyone's judgment, so the flaw that killed LIBOR, a rate that people could shade because it was only an estimate, is gone.

What was given upLIBOR carried information SOFR does not. Because LIBOR measured borrowing on trust, it rose when banks started to doubt each other. In October 2008 it jumped far above the government's own rates, and that gap was one of the clearest alarm signals of the financial crisis. SOFR cannot sound that alarm. A loan backed by Treasury bonds is safe whatever anyone thinks of the borrower, so SOFR does not rise when banks lose faith in each other; in a panic it can even fall, as lenders rush toward safe collateral. Anything in the loop that was using the rate to sense the health of the banks lost that signal, because the new member is not measuring it. The role was filled; it was not filled identically.

ReadingThis is a substitution that keeps the loop closing, and it is better on the exact weakness that caused the old member to fail, but it is blind to something the old member could see. It is neither simply an improvement nor simply a loss. That mixed result is the ordinary case, and it is why asking "is the replacement better?" is too crude a test to carry the weight. The useful questions are what the replacement fixed, what it no longer measures, and whether the loop still closes.

Substitution two: the funding supplier, replaced by rule

Money market funds sit in the liquidity and funding loops as a supplier: they buy the commercial paper and certificates of deposit that fund banks and corporates at short maturities. In 2016 that member was substituted, not by competition, but by a rule with a deadline, the "Oct. 14, 2016, deadline for implementing Securities and Exchange Commission (SEC) reforms".

The scale of the swap is recorded. The Office of Financial Research reported in September 2016 that "Assets of U.S. prime money market funds have decreased by more than $700 billion since the beginning of the year, while assets of government money market funds have increased by about the same amount".

And the effect on the loop those funds were feeding is recorded in the same place. "demand has declined for commercial paper and banks' certificates of deposit (CDs)", and "the lower demand for commercial paper and CDs has increased banks' expenses in raising short-term funds in the interbank market".

Where this page first went wrongAn earlier version said the new member closed the loop, on the grounds that short-term funding continued to exist. That does not follow from the evidence, and the evidence rather points the other way: a government money fund is not permitted to hold commercial paper, so it cannot occupy the position the prime fund vacated. Demand for CP and CDs fell precisely because the replacement does not do that job. Saying the loop closed required naming what actually closed it, and the first version did not.

What actually closed itThe money did not stop funding banks. It was routed through an intermediary that was not previously in the circuit. Gissler and Narajabad, writing for the Federal Reserve Board, record the same episode as a shift of "$1.2 trillion from prime money funds to government money funds" and then report where that money went: "at the end of February 2017, almost one-fifth of the money fund industry's three trillion dollar portfolio was invested in FHLB debt", a holding that by then "stood at more than half of all outstanding FHLB debt". The Federal Home Loan Banks lent the proceeds on: "In January 2017, borrowing by commercial banks comprised over 65 percent of total advances outstanding".

before: prime money funds → commercial paper and CDs → banks
after: government money funds → FHLB discount notes → FHLB advances → banks

Reading, correctedThis is not one member swapped for another. It is a member removed and the circuit re-routed through a new member that had to be added to close it, at a higher cost for the function. The OFR's finding and the FEDS note are consistent: the direct path got more expensive, and a longer path opened beside it.

The proposition still does not survive, and the corrected case makes the point better than the wrong one did. Nothing here rejected the substitution for failing to enhance the loop. Something outside the loop removed a member on a deadline, and the circuit re-formed around the gap in the cheapest way available to it, which was not cheaper than before.

And note who the new member isThe Federal Home Loan Banks are government-sponsored enterprises. The first page's question, which member is handed in from outside, does not go away in this case either. The re-routing ran through one.

Substitution three: the counterparty, in progress right now

The rarest thing on this page is a substitution you can watch rather than reconstruct. In the collateral loop, one member is the party on the other side of the trade. In the US Treasury market that relationship is being restructured around a central counterparty.

And it is novation, not replacementThe original counterparty does not disappear. Through clearing the trade is novated, so that the clearing house stands between the two sides and becomes the legal counterparty to each. The shape is A to CCP to B, not A to CCP with B gone. That is a change in who bears the exposure and in what the circuit routes through, which is why it belongs on this page, but it is intermediation rather than a member being swapped out.

The SEC adopted the rule on 13 December 2023. It requires covered clearing agencies to have policies requiring their members to submit for clearing "all repurchase and reverse repurchase agreements collateralized by U.S. Treasury securities", along with interdealer cash transactions.

The dates moved once, and the current ones are these. The Commission "extended the original compliance dates for the Treasury Clearing Rule by one year to Dec. 31, 2026, for eligible cash market transactions and June 30, 2027, for eligible repo market transactions". As of a statement dated 7 August 2026, no further extension has been announced, and the Commission describes "mandatory clearing for U.S. Treasury cash transactions at the end of the year and U.S. Treasury repo transactions by June 30, 2027".

ReadingA live loop is being restructured in public, on a timetable, with the old and new arrangements documented. The distinction worth drawing from that is not that a financial set is mutable and a chemical one is not: RAF theory accommodates alternative catalysts and alternative pathways, which is the very fact the top of this page relies on. The distinction is authorship. A chemical set's alternatives are discovered by the chemistry. This one was drafted, published for comment, adopted on a date and given a compliance schedule.

The substitute that grows an economy around itself

The second half of the proposition this page was written to examine is the more interesting one: that a member of an established loop can start to develop its own economy from the position it occupies, and to become more efficient by feeding on the larger circuit that sustains it.

The clearing case shows the shape of it. A central counterparty does not merely stand between two trades. It runs a membership, sets its own admission standards, collects initial and variation margin, maintains a default fund, prices its own services, and produces data that only it holds. The circuit it joined supplies all of that, in volume, continuously. None of it existed before the member did.

The consequence is measurable and it has been measured. The FSB, CPMI, IOSCO and BCBS reported in 2018 that "Prefunded financial resources are concentrated at a small number of CCPs", that "Exposures to CCPs are concentrated among a small number of entities", and that "Clearing members and clearing member affiliates are also important providers of other critical services required by CCPs and can maintain several types of relationships with multiple CCPs simultaneously".

That last sentence is the one worth sitting with. The members of the loop supply the substitute with the services it needs to do its job, and the substitute supplies the members with the service that lets them keep trading. A sub-circuit has formed inside the larger one, and it has its own food supply, its own products and its own dependencies.

A hypothesis, labelled as oneThe concentration and interdependency findings above are measurements. What follows from them here is not. The proposal is that a member which builds an economy around its position becomes harder to substitute in turn, because the cost of replacing it is no longer the cost of the function it performs but the cost of the function plus everything built on top, so the loop's ability to swap that member out falls as the member succeeds. The cited reports do not demonstrate that. They establish the dependencies; the claim about rising replacement cost is an inference from them and is offered for testing, not as a finding.

And it moves the question rather than answering itThe first page asked which member of each loop is supplied from outside, on the grounds that the answer decides whether a recovery belonged to the market. A substitution that replaces a fragile member with a robust institution does not retire that question. It gives it a new address. The supplied member is now the clearing house, its rulebook and whoever stands behind it.

What would show this reading wrong

A substitution that arose inside a loop, chose itself, and displaced an incumbent without a rule, a regulator or a standards committee installing it. That would be the loop producing its own member, and it would move these cases from maintenance toward reconstitution.

A documented case of a replacement that closed the circuit, was materially worse for the members, and was expelled by the market rather than repealed by an authority. That would rescue the strong form of the proposition, that the attractor rejects what does not enhance it.

Evidence that the 2016 money fund substitution did not raise short-term funding costs, or that the effect reversed quickly and the reading of it here is an artifact of a few months of data.

A central counterparty that became easier to replace as it grew, rather than harder. That would break the hypothesis that building an economy around a position reduces the loop's ability to swap the position's occupant.

A worked case in which a member was removed, the vacated role was filled by something already in the set, and the circuit still failed. That would confirm the correction made at the top of this page, that filling the role is necessary and not sufficient, with an example rather than an argument.

A study this page suggests and does not run

Every substitution named here has a date, and each replaced member has a measurable function. Collect the substitutions in one loop over several decades, code each for whether it was installed by rule or adopted by choice, and for what happened to the cost of the function afterward. The question is whether chosen substitutions and imposed substitutions leave different signatures, and whether either predicts how the loop behaves at the next shock.

Nothing of that is done here. It is named because the page makes a claim about how substitutions get admitted, and that claim is checkable against a list somebody could assemble.

Honest limits

This is a reading of published work. Nothing on this page was measured, modelled or simulated here, and there is no panel.

Autocatalysis is used by analogy throughout, as on the first page. None of these loops meets the chemical definition, where the catalysts are molecules and the closure is exact. The RAF condition is quoted because it states precisely what substitution requires, not because a market satisfies it formally.

The word member is carrying three different kinds of thing, and the page should say so rather than let the vocabulary smooth it over. A reference rate is a number that contracts read. A class of money market fund is a type of intermediary. Central clearing is a piece of market architecture. RAF theory does not license treating those as the same kind of object, and the only thing they genuinely share here is that each occupies a position the circuit depends on. Anyone pushing on this page should push there first.

Three substitutions are not a sample. They were chosen because each is documented with dates and because each sits in a loop the first page already wrote out. A wider set might not behave like these three, and the second falsifier above is the reason the strong form of the proposition is reported as unsupported rather than as refuted.

The Treasury clearing dates are live and have already moved once. They were read on 20 September 2026 and could move again.

No sovereignty verdict, no metric, no kernel construct, and no investment advice.

Sources

Hordijk, W., & Steel, M. (2018). Autocatalytic Networks at the Basis of Life's Origin and Organization. Life, 8(4), 62. Source of the RA and F conditions and of the "at least one molecule type" wording the substitution argument rests on. doi.org/10.3390/life8040062

Alternative Reference Rates Committee, SOFR transition pages, Federal Reserve Bank of New York. Source of the 2017 selection of SOFR, the 30 June 2023 cessation of USD LIBOR panel settings, the description of SOFR as a Treasury repo rate, and the statement that LIBOR was not anchored in market activity. newyorkfed.org/arrc/sofr-transition

Alternative Reference Rates Committee (2018, March). Second Report. Federal Reserve Bank of New York. Source of the estimate of roughly $200 trillion in contracts referencing USD LIBOR. Added 2 October 2026.

Federal Reserve Bank of New York, Secured Overnight Financing Rate data and methodology. Source of the start of SOFR publication on 3 April 2018 and of the daily transaction volume underlying it. newyorkfed.org/markets/reference-rates/sofr. Added 2 October 2026.

U.S. Commodity Futures Trading Commission, U.S. Department of Justice and U.K. Financial Services Authority (2012, June 27). Barclays settlement over LIBOR submissions (about $450 million in combined penalties). Added 2 October 2026.

Sengupta, R., & Tam, Y. M. (2008). The LIBOR-OIS spread as a summary indicator. Economic Synopses, No. 25, Federal Reserve Bank of St. Louis. Source of LIBOR's jump above government rates in October 2008 as a stress signal. Added 2 October 2026.

Office of Financial Research (2016, September 22). OFR Monitor Shows Accelerating Shift to Government Money Market Funds. Source of the $700 billion shift, the 14 October 2016 compliance deadline, and the effect on commercial paper, certificates of deposit and bank short-term funding costs. financialresearch.gov, OFR blog

Gissler, S., & Narajabad, B. (2017, October 18). The Increased Role of the Federal Home Loan Bank System in Funding Markets, Part 2: Recent Trends and Potential Drivers. FEDS Notes, Board of Governors of the Federal Reserve System. Source of the $1.2 trillion shift, the money fund industry's FHLB debt holdings at February 2017, the share of outstanding FHLB debt held by money funds, and the commercial bank share of advances in January 2017. federalreserve.gov, FEDS Notes

U.S. Securities and Exchange Commission (2023, December 13). Press release 2023-247, on standards for covered clearing agencies for U.S. Treasury securities. Source of the adoption date and the scope of the clearing requirement. sec.gov press release 2023-247

Uyeda, M. (2026, August 7). Update on the SEC's Work Toward Treasury Clearing Implementation. Source of the current compliance dates, 31 December 2026 for cash and 30 June 2027 for repo, and of the one-year extension dated 25 February 2025. sec.gov, August 2026 statement

FSB, CPMI, IOSCO and BCBS (2018, August 9). Analysis of Central Clearing Interdependencies. Source of the concentration and interdependency findings. fsb.org

Related on this site: The Autocatalytic Sets of the Financial Markets, which writes the four loops out; The Crash of 2008; and The Bank Run.