The Autocatalytic Sets of the Financial Markets
Four loops that make more of themselves, written out member by member, with the outside supply each one runs on and the dated occasion when that supply was cut.
A market is one of the few large systems where a self-feeding loop can be written down part by part, because every part leaves a record. Each of the four loops below is already documented in the economics literature under its own name. What this page adds is the question the rest of this section keeps asking: which member of the loop is produced inside it, and which one is handed in from outside, because the answer decides whether a recovery belonged to the market or to whoever supplied the missing part.
Each loop is set out the same way: the members, the circuit they form, what comes in from outside, the member whose loss stops it, and an episode with a date where that member was in fact removed or withheld.
Every figure and every quotation is sourced in the list at the bottom, with the date it was read. The loops are not this site's discovery; the sources name them. Nothing here is investment advice.
On this site a basin is the set of conditions a system comes back from. Push a marble around the inside of a bowl and it rolls back to the bottom; the bowl is its basin, and the rim is its edge, the push beyond which it does not come back. Finding basins, and their edges, is the point of the Laboratory. The four loops below are four candidate basins, each held in place by an outside supply, and the edge of each is the dated occasion that supply was cut. This is a reading page with no panel, so the pushes are the ones history supplied.
| basin | what it returns to | what pushes it | its edge |
|---|---|---|---|
| 1. Collateral and leverage | credit funding purchases that hold up the collateral | a fall in price | haircuts raised, 2007 to 2008 |
| 2. Liquidity | prices back in line within days | trading flows larger than dealers can hold | dealers overwhelmed, March 2020 |
| 3. Deposits | nobody withdrawing early | withdrawals others can see | the queue meaning something; insurance removed that meaning, 1 January 1934 |
| 4. Flows | unclear | inflows after good performance | no dated cut on record |
1. Collateral and leverage. It returns to credit, purchases and marked value feeding one another. The member supplied from outside is the haircut rule (the share of a pledged asset's value a lender will not lend against), set by risk managers and regulators. A large enough fall in price runs the loop backward as margin calls and forced selling. The edge: 2007 to 2008, when Gorton and Metrick's haircut index rose "from zero in early 2007 to nearly 50% at the peak of the crisis in late 2008," and borrowing capacity fell with no change in the assets.
2. Liquidity. A price knocked sideways on a Tuesday is usually back by Thursday, because volume pays dealers to keep quoting. The outside supply is dealer funding. The edge: March 2020 in Treasuries, when "dealers were overwhelmed by the trading flows." The market came back because the Federal Reserve bought $1.77 trillion of Treasuries and $892 billion of agency mortgage-backed securities between 13 March and 31 July 2020.
3. Deposits. Diamond and Dybvig showed the same arrangement has two outcomes: the ordinary one, where nobody withdraws early, and the run. Visible withdrawals push it toward the run. The edge: the point where the queue starts to mean that waiting is risky. Federal deposit insurance took that meaning away on 1 January 1934, and "only nine banks failed in 1934, compared to more than 9,000 in the preceding four years." The member cut was the run's catalyst, so cutting it deepened the ordinary basin.
4. Flows may have no basin. Lou finds that fund inflows push the prices of stocks the funds already hold, fed by household savings and index rules. Flows reverse and the effect is documented as temporary, so nothing shows that the loop holds anything in place.
Basin depth, in "Signs a loop is turning inward" below, is how large a shock it takes to move a loop at all. In the first three loops something made outside the loop sets that depth: the haircut rule, dealer funding, deposit insurance.
What an autocatalytic set is, in plain terms
In chemistry, a set of molecules is called autocatalytic when every member of the set is produced by a reaction that some other member speeds up, so the set as a whole keeps making itself. The refinement that matters here is that the set is fed: raw materials arrive from outside and are not produced by the loop. A set that keeps making itself from a supply of raw food is called a RAF set, and the two halves of that idea, closure on the inside and food from outside, are exactly what this page is looking for in a market.
The parallel is a parallel, and the page says so plainly. A haircut rule is not a molecule. What carries over is the structure: a part that speeds the production of another part without being used up in the process, a circuit that closes, and a supply line that does not. What also carries over is the test, which is to name the member you think is doing the catalysing and find the occasion when it was absent.
Set one: collateral and leverage
This is the one most often described, under several names: the leverage cycle, the financial accelerator, funding and market liquidity spirals. The members are ordinary and none of them is hidden.
- A traded asset with a quoted price. The price is produced inside the loop, by trades.
- A valuation convention that turns the price into a number on a balance sheet: marking to market, pricing services, ratings.
- A haircut or margin rule that converts the marked value into borrowing capacity.
- A lender willing to extend credit against that capacity.
- Buyers who use the credit to buy more of the asset.
- A dealer who produces the price by standing ready to trade, and whose willingness depends on its own funding.
purchases raise the price → the price raises the marked value → the marked value raises borrowing capacity → credit funds more purchases
Fed from outsidesavings and income earned elsewhere, enforceable contracts and custody, the payment system, and the willingness of a central bank to lend against the same collateral when no one else will.
The catalysisthe haircut rule is the clearest case. It is not consumed by the lending it enables, and changing it changes the rate at which credit is produced without changing anything about the asset. In the chemical picture that is what a catalyst does.
Runs backward toothe same members with the sign reversed give the deleveraging spiral: falling price, falling marked value, margin calls, forced selling, falling price. A loop that runs both ways is better evidence that the loop is real than any single direction on its own.
The episode2007 to 2008. Gorton and Metrick construct a repo haircut index and report that it "rises from zero in early 2007 to nearly 50% at the peak of the crisis in late 2008." That is member three being turned off. Borrowing capacity fell without any change in the assets themselves, and the selling that followed lowered the prices that set the capacity.
Where it already lives on this sitethe housing version of this loop, told as a history, is The Crash of 2008.
Set two: liquidity
The first set explains booms and busts. This one explains something quieter and more common: why a price knocked sideways on a Tuesday is usually back in line by Thursday.
- Traders who want to transact and can only do so if someone takes the other side.
- Dealers and market makers who take that side, holding inventory and funding it.
- The spread and depth they quote, which is the price of transacting.
- Volume, which is what makes the dealer's business worth doing.
volume → dealers earn enough to keep quoting → spreads stay narrow → transacting stays cheap → volume
Fed from outsidedealer funding, which is where this set touches the first one. Brunnermeier and Pedersen set out the connection between the liquidity of a market and the funding available to the people who provide it, and the joint spiral that follows when both tighten together.
The breakthe dealer. Remove the willingness to hold inventory and the price stops being produced, whatever the asset is worth. A thinly traded stock that moves five percent on a single order is this loop at low amplitude.
The episodeMarch 2020, in the largest and supposedly deepest market there is. A Federal Reserve Bank of New York staff report records that "dealers were overwhelmed by the trading flows" and that "their balance sheet constraints and internal risk limits prevented them from meeting the increased liquidity demand," with market liquidity deteriorating "to its worst level since the GFC."
Who supplied the returnthe Federal Reserve, and at a scale that is worth stating precisely. Cumulative purchases between 13 March and 31 July 2020 came to $1.77 trillion of Treasuries and $892 billion of agency mortgage-backed securities, made under an instruction to buy "in the amounts needed to support the smooth functioning" of those markets. The market came back. The member that had gone missing was handed in from outside.
Set three: deposits
A bank run is the cleanest self-feeding loop in finance, and the only one on this page that a government took apart on purpose.
- Depositors who can withdraw on demand.
- A bank holding illiquid loans against those demandable deposits.
- The queue, meaning the visible fact of other people withdrawing.
- The belief that the bank may not have enough to pay everyone.
withdrawals → a visible queue → the belief that waiting is risky → more withdrawals
Fed from outsidethe assets, the legal claim to them, and the arrangement that decides who is paid first. Diamond and Dybvig showed that the run is one of two possible outcomes of the same arrangement, with the other being the ordinary state in which nobody withdraws early.
The catalyst, and its removalthe queue's meaning. Deposit insurance does not stop people withdrawing; it makes the queue stop meaning anything, because the money is guaranteed whether or not you are near the front. Federal deposit insurance became effective on 1 January 1934. The FDIC's own history records that "only nine banks failed in 1934, compared to more than 9,000 in the preceding four years."
Still the sharpest demonstration on this pageone member was taken out of the loop by law, on a known date, and the phenomenon the loop produced very largely stopped. That is as close to a controlled experiment as this subject offers.
Where it already lives on this sitethe simulation is The Bank Run.
Set four: flows
- A fund holding a set of positions.
- Its recent performance, which is what savers look at.
- Inflows from those savers.
- Purchases of the same positions the fund already holds.
performance → inflows → buying the existing holdings → performance
Fed from outsidehousehold savings, which the loop plainly does not produce, and the index rules that decide what gets bought.
The evidenceLou finds that fund flows push the prices of the stocks those funds already hold, which is the mechanical part of the circuit. The step from that to a self-sustaining loop is where the claim gets weaker, because flows reverse and the effect is documented as temporary.
Why it is here anywaythe page would be dishonest if it showed only the loops that work. This one has a real mechanism and an unclear closure, which is the ordinary condition of most candidate loops anyone will find.
Which member comes from outside
Every loop above is real and every one of them is fed. The interesting line is not between loops that self-feed and loops that do not. It is between loops that could remake their own members and loops that depend on a member nobody inside them produces.
| Loop | Member supplied from outside | What happened when it was absent |
|---|---|---|
| Collateral and leverage | The haircut rule, set by risk managers and regulators | Haircuts to nearly 50% by late 2008; borrowing capacity collapsed with no change in the assets |
| Liquidity | Dealer funding and balance sheet capacity | March 2020: price production stopped in Treasuries until the central bank bought $1.77 trillion of them |
| Deposits | The meaning of the queue, which insurance removed | Nine bank failures in 1934 against more than 9,000 in the four preceding years |
| Flows | Household savings and index rules | Flows reverse and the price effect is documented as temporary |
Read the second column and a pattern appears. In three of the four, the part that is not produced inside the loop is a rule or a public institution. That is not a complaint about markets. It is a statement about where the return comes from, and it is checkable, which is more than most statements about markets manage.
Signs a loop is turning inward
A loop can grow harder to leave and more expensive to change without growing any larger. That is the condition worth naming: not a bigger market, but one where more of the parts are wired to each other, where the same risk model runs everywhere, and where the rules that bound the loop are increasingly written by the people inside it.
The framework this site uses has a construct for exactly that resistance, logic mass, and it is not available here. Its definition restricts it to "sovereign attractors during their bootstrapping interval," and it is "undefined for: non-sovereign configurations, attractlets, pre-ignition states, and post-β-loss states." Every loop on this page is fed from outside, which puts all four on the attractlet side. Calling any number below a logic mass would break this site's own rules, so none of them is called that.
What the framework does allow is the list of things that estimate it: "interlock density
measurements, basin-depth estimators, T₃/T₄ historical cost integrals, and
α-trace richness measures," together with the caution that "metrics estimate
mᵧ; metrics do not define logic mass." Those four translate into
market observables without much strain.
- Interlock density. How many members now depend on one another with no slack left. Collateral reused down longer chains, cross-margin agreements, clearing concentrated in one or two houses, everyone pricing from the same vendor.
- Basin depth. How large a shock it now takes to move the thing at all. The tell is falling dispersion: when every risk system says the same thing, small shocks do nothing and the shock that matters moves everyone on the same afternoon.
- Cost of reconfiguration. What changing it would take. Standard documentation across the whole market, plumbing built for one workflow, firms and careers specialized to it.
- Stored history. Precedent, carried in prices rather than in anyone's statement. Each remembered rescue lowers the cost of the next round of leverage.
The general signature, across all four loops: activity whose only purpose is servicing the loop grows faster than the activity the loop was built for.
| Loop | What turning inward looks like | Where the data lives |
|---|---|---|
| Collateral and leverage | Collateral reuse chains lengthening; growth in collateral transformation, a business that exists only to feed the haircut member; haircut dispersion falling across lenders; term funding giving way to overnight | Financial Stability Board non-bank intermediation monitoring; Office of Financial Research repo collections; FINRA margin debt |
| Liquidity | The wrapper becoming the venue, with fund volume growing against volume in the underlying; market making concentrated in a few firms; dealer hedging of options becoming a driver of the thing being hedged, which is the loop closing on itself | Exchange volume reports; Options Clearing Corporation volumes; fund creation and redemption data |
| Deposits | Uninsured share rising; deposits concentrating in fewer institutions; the withdrawal channel getting faster, so a queue that took three days now takes an afternoon | FDIC call report data on insured and uninsured deposits |
| Flows | Rule-driven buying growing against view-driven buying: index inclusion, target-date rebalancing, volatility targeting; benchmark monoculture, where the index provider's decision is the reason the asset is held | Passive share of assets under management; published index rebalance calendars |
None of these is a threshold and none of them is a verdict. They are directions of travel, and a loop moving in all four at once is the case worth looking at closely.
Predicting behavior under perturbation
The obvious use of a loop you have written out is to say what it will do when something hits it. That turns out to be three separate claims, and lumping them together is how people get hurt. Nothing here is investment advice, and none of it is a prediction about any market.
One, mechanical flowsSome responses follow from published rules and are therefore knowable in advance: index rebalances, margin calls at stated thresholds, funds that target a level of volatility and must sell after a spike, redemption gates, quarter-end balance sheet effects. The direction and the timing come from the rulebook. This part is real, it is also thoroughly worked over by people with faster systems, and the edge in it decays as more of them arrive.
Two, fragility indicatorsEverything in the section above, plus the slowing recovery and rising variance that the field guide describes, tells you about the shape of the distribution rather than the date. In markets these have a long record of being right eventually and useless in the meantime. They are better suited to deciding how much to hold than to deciding when to act.
Three, the supplier questionThis is the one this page adds. When a loop breaks, ask whether the member it cannot produce has a supplier who is both able and committed. In March 2020 the answer was yes, and the size of the answer was $1.77 trillion. For mortgage credit to marginal borrowers after 2008, the answer was no: the driver went back to zero and the behavior did not return. The same question separates drawdowns that come back from drawdowns that do not, and it is a judgment about institutions rather than a reading of a chart.
The hazard, stated plainlyA position built on the return is short the tail. It earns steadily while the supplier keeps showing up and gives all of it back on the occasion when the supplier is absent, constrained or outvoted. The people who were carried out in 1998 and in early 2018 were not wrong about the loop. They were wrong about one instance of the supplier, which was the only instance that mattered.
A study rather than a signal
The supplier question can be tested on public data, and testing it is more useful than trading it.
- Assemble episodes. Drawdowns of comparable size across markets and decades, with dates.
- Code each one for the member that went missing, whether it was restored, by whom, how fast, and at what scale.
- Record the outcome: returned to the old level, returned to a different level, or never returned, with the recovery time where there was one.
- Test whether the supplier coding predicts the outcome better than drawdown size alone.
Predictions and the conditions that would prove them wrong get written down before the data is touched, in the same way as the other studies on this site. A null result would be worth publishing: it would mean the supplier framing adds nothing beyond what the size of the fall already tells you.
Three ways to fool yourself, all of them common here
Price is not the statea price is one output of the system. Saying a market returned to the same level means nothing until you say in what units, over what horizon, and against what growth or inflation.
Reversion is not returna bounded, noisy series drifts back toward its average for statistical reasons that need no loop at all. Markets are where this confusion is most seductive, because the drift back is real and the explanation is usually invented afterward.
Survivorshipan index recovers partly because its losers were removed from it. The index comes back; the companies do not. A recovery measured on a survivor-adjusted series is measuring the adjustment.
There is a fourth, slower problem. The system does not hold still. Rules, participants and technology change, so a recovery time measured across fifty years is a measurement of several different systems wearing the same name.
What would show this reading wrong
For the collateral loop: a period in which haircuts were raised sharply and broadly while leverage and prices held steady would break the causal step from member three to member four.
For the liquidity loop: a market that loses its dealers and continues to produce reliable prices at ordinary spreads.
For the deposit loop: widespread runs on fully insured deposits, where the queue still carries its old meaning despite the guarantee.
For the flow loop: no relation between fund flows and the prices of the holdings those funds already own, which is the finding the loop rests on.
Who got here first
Carrying autocatalysis out of chemistry and into economies has been going on for more than twenty years, by several groups working separately. The closest prior work states the move in almost the words used above.
Padgett, and the production chemistryPadgett, Lee and Collier open their 2003 paper by saying it is "inspired by the hypercycle model of the origins of chemical life on earth" and that it "develops an autocatalytic model of the co-evolution of economic production and economic firms, represented as skills." Their survival condition will be familiar by now: "the minimal requirement for long-term survival, both of firms and of clusters, is to participate in at least one spatially distributed production chain that closes in on itself, to form a loop." That program continued into Padgett and Powell's 2012 book, which applies the same machinery to the emergence of organizations, markets and Renaissance banking. The chemistry reached finance before this page did.
Kauffmanhas argued in his own writing that the real economy is collectively autocatalytic. It is an assertion rather than a model, and it is his.
The autocatalytic set communityhas reached toward economies from the formal side. Hordijk's 2013 review describes "some (still speculative) ideas of how this theory can potentially be applied to living systems in general and perhaps even to social systems such as the economy," and the RAF apparatus has since been applied to technological evolution.
Neighbours with different machineryBak, Chen, Scheinkman and Woodford brought self-organized criticality into production and inventory dynamics in 1993. Ulanowicz, Goerner and Lietaer carried ecological network flow analysis into economic systems in 2009, which is the source of the argument that a system can be too efficient to survive. Same instinct, different apparatus.
And the finance literature itselfdescribes all four loops in detail without ever using the chemistry. Minsky on financial instability, Soros on reflexivity, and the spiral papers cited above have the loops; what they do not have is the closure question.
What is left for this pagea narrow claim, and it is the only one made here. The mapping is old. What could not be found is anyone writing these specific financial loops as member lists with a food set, and then using the supplied-member test against dated episodes. Padgett models production and organizations, Hordijk is explicit that the economic application is speculative, and the finance papers have the mechanisms without the question. So: not the first to map autocatalysis onto economies, possibly the first to set out these four this way and ask which member is handed in from outside.
Honest limits
It is a reading, not a result. No data was gathered or analyzed here; every number is quoted from a published source.
The four loops are named and studied in economics under their own names. The contribution claimed here is the framing question about supplied members, and that framing is an interpretation rather than a finding.
Autocatalysis is used by analogy. None of these sets meets the chemical definition, in which the catalysts are molecules and the closure is exact.
Nothing here is investment advice, and none of it is a prediction.
Sources
Brunnermeier, M. K., & Pedersen, L. H. (2009). Market Liquidity and Funding Liquidity. Review of Financial Studies, 22(6), 2201–2238. academic.oup.com/rfs/article-abstract/22/6/2201/1592184
Geanakoplos, J. (2010). The Leverage Cycle. In NBER Macroeconomics Annual 2009, Volume 24, University of Chicago Press, pp. 1–65. nber.org/books-and-chapters/nber-macroeconomics-annual-2009-volume-24/leverage-cycle
Bernanke, B. S., Gertler, M., & Gilchrist, S. (1999). The Financial Accelerator in a Quantitative Business Cycle Framework. Handbook of Macroeconomics, Volume 1, Chapter 21, pp. 1341–1393. sciencedirect.com/science/chapter/handbook/pii/S157400489910034X
Adrian, T., & Shin, H. S. (2010). Liquidity and Leverage. Journal of Financial Intermediation, 19(3), 418–437. Working paper version: newyorkfed.org/medialibrary/media/research/staff_reports/sr328.pdf
Gorton, G., & Metrick, A. (2012). Securitized banking and the run on repo. Journal of Financial Economics, 104(3), 425–451. Source of the haircut index quotation. nber.org/papers/w15223
Shleifer, A., & Vishny, R. (2011). Fire Sales in Finance and Macroeconomics. Journal of Economic Perspectives, 25(1), 29–48. aeaweb.org/articles?id=10.1257/jep.25.1.29
Diamond, D. W., & Dybvig, P. H. (1983). Bank Runs, Deposit Insurance, and Liquidity. Journal of Political Economy, 91(3), 401–419. journals.uchicago.edu/doi/10.1086/261155
Federal Deposit Insurance Corporation, A Brief History of Deposit Insurance in the United States. Source of the 1 January 1934 effective date and the 1934 failure comparison. fdic.gov/resources/publications/brief-history-of-deposit-insurance
Adrian, T., Fleming, M., & Nikolaou, K. (2025). U.S. Treasury Market Functioning from the GFC to the Pandemic. Federal Reserve Bank of New York Staff Report No. 1146. Source of the dealer-constraint quotations. newyorkfed.org/medialibrary/media/research/staff_reports/sr1146.pdf
Garbade, K. D., & Keane, F. M. (2020). Market Function Purchases by the Federal Reserve. Liberty Street Economics, Federal Reserve Bank of New York, 20 August 2020. Source of the $1.77 trillion and $892 billion figures and the quoted instruction. libertystreeteconomics.newyorkfed.org/2020/08/market-function-purchases-by-the-federal-reserve
Lou, D. (2012). A Flow-Based Explanation for Return Predictability. Review of Financial Studies, 25(12), 3457–3489. academic.oup.com/rfs/article-abstract/25/12/3457/1594242
Padgett, J. F., Lee, D., & Collier, N. (2003). Economic production as chemistry. Industrial and Corporate Change, 12(4), 843–877. Source of the hypercycle and production-loop quotations. academic.oup.com/icc/article/12/4/843/845803
Padgett, J. F., & Powell, W. W. (2012). The Emergence of Organizations and Markets. Princeton University Press. press.princeton.edu/books/paperback/9780691148878
Hordijk, W. (2013). Autocatalytic Sets: From the Origin of Life to the Economy. BioScience, 63(11), 877–881. Source of the quoted “still speculative” passage. academic.oup.com/bioscience/article/63/11/877/2389920
Kauffman, S. (2011). Economics and the Collectively Autocatalytic Structure of the Real Economy. NPR, 21 November 2011. npr.org/sections/13.7/2011/11/21/142594308
Emergence of autocatalytic sets in a simple model of technological evolution. Journal of Evolutionary Economics (2023). link.springer.com/article/10.1007/s00191-023-00838-2
Bak, P., Chen, K., Scheinkman, J., & Woodford, M. (1993). Aggregate fluctuations from independent sectoral shocks: self-organized criticality in a model of production and inventory dynamics. Ricerche Economiche, 47(1), 3–30. sciencedirect.com/science/article/abs/pii/003550549390023V
Ulanowicz, R. E., Goerner, S. J., Lietaer, B., & Gomez, R. (2009). Quantifying sustainability: Resilience, efficiency and the return of information theory. Ecological Complexity, 6(1), 27–36. sciencedirect.com/science/article/abs/pii/S1476945X08000561
On autocatalytic and RAF sets, this site's background pages: Autocatalytic Sets and the Origin of Life and RAF Sets and Catalytic Closure.