Banking And Finance Closed Loop Systems | The Complete System From Deposits, Credit and Payments to Settlement, Risk and World Return

Banking and finance closed loop systems are a way to understand the banking system and the wider financial system as connected cycles of deposits, loans, money creation, payments, clearing and settlement, liquidity, bank capital, risk, financial market infrastructure and real-world cash flows. Instead of treating a bank account, loan, payment system, balance sheet or capital ratio as an isolated topic, this guide follows the full return path: where a financial claim begins, how it moves, what must settle, what can fail, how risk changes the next decision, and whether money and credit return as repayment, income, loss or renewed capacity.

That full-system view captures many of the search questions readers actually ask—how banking works, how the financial system works, how banks create money, how payment systems move money, what clearing and settlement mean, why banks need liquidity and capital, how systemic risk spreads, and how financial stability is maintained. It also separates two meanings of “closed loop” that are often confused. A closed-loop payment product is usually a restricted payment network in which value circulates among approved users or merchants; a closed-loop systems analysis is broader, asking whether the information, money, obligations, risk and feedback generated by one stage return to change the next stage.

For readers in Singapore and Bukit Timah, the same systems logic can be traced through everyday services such as bank deposits, cards, FAST and PayNow, and onward into clearing, settlement and central-bank infrastructure. The Monetary Authority of Singapore currently lists FAST, Inter-bank GIRO, the Singapore Dollar Cheque Clearing System, NETS EFTPOS and the New MAS Electronic Payment and Book-Entry System among designated payment-system activities, while the Association of Banks in Singapore explains that PayNow transfers Singapore-dollar funds through FAST. The local rails are specific; the mathematical questions—stocks, flows, delays, queues, network links, buffers, feedback and failure propagation—are universal.

Reader boundary. This is an applied-mathematics and systems-thinking article. It is not personal financial advice, investment advice, banking-operating guidance or a claim that Bukit Timah Tutor is a banking authority. Financial, legal and regulatory facts should be checked against the competent institution. Bukit Timah Tutor’s owned job here is to make the mathematics of the system visible.

The 50-second router

• If you want the one-line model, read The master loop.
• If you want to know how a loan can create a deposit, go to The credit-creation loop.
• If you want to know why “payment sent” is not the same as “settlement final”, go to The payment, clearing and settlement loop.
• If you want bank survival mathematics, read Liquidity, capital and solvency are different constraints.
• If you want the Singapore layer, read Singapore: from PayNow and FAST to final settlement.
• If you want equations, stocks-and-flows, networks and feedback, read The mathematical toolkit.
• If you want failures, read How loops break and the diagnostic atlas.
• If you want the specialist mechanism library, use Finance & Banking Algorithms | Applied Mathematics in Real Financial Systems.

The master loop: finance only becomes intelligible when the return path is visible

A useful master loop is: income or funding → deposit or financial claim → balance-sheet position → credit or investment decision → payment instruction → clearing → settlement → real-world use → cash flow or asset-value change → repayment, return or default → profit or loss → capital and liquidity update → risk assessment → new decision. The sequence is not a literal description of every product. It is a systems map. It forces a reader to ask what enters, what is stored, what leaves, what has not yet settled, what feedback arrives later, and what variable changes the next cycle.

The map has several layers at once. The accounting layer records assets, liabilities, equity, income and expense. The payment layer moves instructions and settlement assets. The credit layer creates and monitors claims. The risk layer estimates uncertainty and potential loss. The market layer changes prices and funding conditions. The regulatory layer imposes constraints and recovery requirements. The real-economy layer determines whether funded activity produces wages, sales, taxes, rents, output or losses. A bank can look healthy on one layer while a different layer is under stress, which is why a closed-loop view should never compress everything into one number.

A loop is “closed” analytically when the consequences of an action return as information or resources that influence the next action. A mortgage decision does not end when the loan is booked. It changes deposits, payments, settlement needs, interest income, credit exposure, liquidity use, capital requirements and eventually default or repayment data. Those outcomes influence future underwriting, pricing, provisioning and balance-sheet capacity. The return path is what turns a transaction into a system.

The idea also exposes false endings. “The customer paid” may mean the payer authorised an instruction, not that interbank settlement is final. “The bank has capital” does not mean it has cash available at the instant depositors withdraw. “The bank has liquidity” does not mean its assets exceed its liabilities after losses. “The loan was repaid” closes one contractual claim but can open another cycle if principal is lent again. Precise systems language prevents one stage from masquerading as the whole process.

Why this lane exists beside the existing Bukit Timah Tutor finance-and-banking mathematics library

Bukit Timah Tutor already has specialist articles on amortisation, credit scoring, liquidity stress, risk-weighted assets, interbank networks, settlement, reconciliation, margin, foreign-exchange settlement and many other mechanisms. Those pages own individual mathematical jobs. This lane owns a different job: how the mechanisms fit together as feedback systems. It therefore uses existing specialist pages as components rather than rewriting them under new names.

For example, the article on payment systems, graphs, queues and real-time settlement can explain a network mechanism in detail. The closed-loop article asks what the payment does to deposit balances, reserves, intraday liquidity, reconciliation queues and subsequent risk decisions. Likewise, bank capital mathematics owns risk-weighted assets and capital ratios; the closed-loop view asks how losses, retained earnings, balance-sheet growth and risk-weight changes feed back into lending capacity.

The distinction matters for search quality as well as conceptual quality. A reader looking for “how payment systems move money” deserves a focused answer. A reader searching “how banking works as a system” needs the relations among payments, deposits, credit, liquidity, capital and feedback. Building separate owners around separate reader jobs reduces cannibalisation and lets each page go deeper without pretending every query is the same query.

Closed loop does not mean sealed system

In engineering, a closed-loop controller measures the state of a process, compares it with a target and adjusts an input. Banking and finance contain many feedback structures with that shape, but they are not sealed machines. They are open to households, firms, governments, markets, laws, technology, expectations and shocks. “Closed loop” here means that the analysis follows the return of consequences, not that the financial system is isolated from society.

The outside world matters because most financial claims are promises about future external events. A business loan is repaid from future cash flow generated by customers. A mortgage depends on household income and property-related obligations. A government bond depends on public finance. A share derives value from expected future enterprise cash flows. Insurance claims depend on events outside the financial contract. Finance is therefore a claim system whose truth is repeatedly tested by the world.

This is also why purely circular explanations fail. A bank cannot ultimately validate its loan book merely by pointing to its own accounting entries. Accounting records the claim; the borrower’s ability to produce or obtain money over time determines whether the claim performs. The strongest closed-loop model therefore includes a world-return stage: funded activity leaves finance, interacts with the real economy, and returns as cash flow, value change, repayment or loss.

Accounting identity: the first invariant

The basic balance-sheet identity is Assets = Liabilities + Equity. It is not a forecasting model or a proof of safety. It is an accounting relationship. If a bank makes a loan and credits a deposit, both an asset and a liability can rise. If a borrower repays principal from a deposit at the same bank, the loan asset and deposit liability can both fall. If an asset suffers a loss that is recognised, equity can fall. Closed-loop reasoning uses the identity as a conservation-style constraint on recorded positions.

A common error is to treat the balance sheet like a box of money. It is a structured list of claims, obligations and residual ownership. A loan asset is not cash; it is a claim on a borrower. A deposit is not an asset of the bank; it is a liability of the bank to the depositor. Cash and central-bank reserves are assets. Equity is the residual after liabilities are subtracted from assets. This classification is more than bookkeeping because different items behave differently under payment pressure, market moves and loss.

The identity also makes feedback visible. Suppose credit losses reduce assets by 10 while liabilities are unchanged. Equity falls by 10. If regulation or internal policy requires a minimum capital ratio, the bank may then retain more earnings, issue equity, reduce distributions, change asset composition, tighten underwriting or slow balance-sheet growth. A loss today changes tomorrow’s feasible action set. That is a genuine control loop.

Stocks and flows: why bank statements and payment volumes answer different questions

A stock is measured at a point in time; a flow is measured over an interval. Deposits outstanding at midnight are a stock. Payments processed during the day are a flow. Bank capital at quarter-end is a stock. Net income over the quarter is a flow that can change equity. New loan originations are a flow; the loan book outstanding is a stock. Confusing stocks and flows produces some of the most persistent errors in finance.

One can formalise the relationship with a simple difference equation: Stock(t+1) = Stock(t) + Inflows − Outflows + Revaluations + Other Adjustments. For deposits, inflows can include incoming transfers, salary credits or newly created deposits linked to lending; outflows include transfers, cash withdrawals and certain repayments. For capital, retained earnings can add to equity while losses and distributions can subtract from it. For liquidity buffers, purchases and inflows add resources while settlements and withdrawals consume them.

The equation is deliberately generic. Its power lies in forcing a complete reconciliation. If a balance changed, what flow or revaluation caused the change? If a payment volume was enormous but end-of-day reserves barely changed, was there offsetting flow, netting or liquidity recycling? If profits were positive but capital fell, were there valuation losses, distributions or other deductions? Closed-loop analysis is strongest when each stock has a corresponding flow story.

The deposit loop: a deposit is both a customer asset and a bank liability

From a customer’s viewpoint, a bank deposit is an asset: a claim that can normally be transferred, withdrawn or used to make payments. From the bank’s viewpoint, the same deposit is a liability. This two-sided relationship is the beginning of a useful systems map. A deposit arrives, changes the bank’s funding composition and payment obligations, may support broader balance-sheet activity, can leave through a payment, and can return through income, transfers or new deposits.

The deposit loop has behavioural feedback. Depositors react to interest rates, service quality, convenience, perceived safety and competing products. Banks observe deposit stability, rate sensitivity and concentration, then change pricing and liquidity planning. A rate increase intended to retain deposits raises funding cost. Higher funding cost can affect loan pricing or profitability. Loan pricing can affect demand and credit quality. The loop therefore reaches far beyond the savings-account screen.

Digital banking can shorten feedback delay. A depositor can move funds in seconds instead of visiting a branch. Faster action is not automatically destabilising; fast payments can create enormous social value. But the time constant of behaviour matters. If withdrawal decisions occur faster than asset sales, collateral mobilisation, funding replacement or management response, a liquidity loop can become more fragile. Systems mathematics therefore asks not only “how much?” but “how fast?”

The credit-creation loop: loan, deposit, spending, repayment and loss

When a commercial bank makes many kinds of loans, it typically records a loan asset and credits a deposit liability. The Bank of England has explained this mechanism in its work on money creation: lending can create a matching bank deposit rather than merely handing over pre-existing notes. That does not mean lending is unconstrained. Capital, liquidity, risk appetite, borrower quality, regulation, funding conditions and expected profitability all shape what the bank can and will do.

The newly created deposit may quickly leave the originating bank when the borrower spends. If the seller banks elsewhere, the payer’s bank and receiver’s bank must settle the interbank obligation through the relevant infrastructure. Credit creation can therefore produce a later liquidity requirement. The loan asset may remain on one bank’s balance sheet while the deposit created by the loan migrates to another bank. That separation is central to understanding why “banks create deposits” and “banks need funding and liquidity” are not contradictory statements.

The loop closes over time through interest and principal payments, refinancing, restructuring, recovery or default. Performing loans generate contractual cash flows. Defaults generate losses after collateral and recoveries are considered. Those results update expected-loss models, underwriting rules, pricing, provisions, capital and portfolio limits. Good credit systems learn from realised outcomes while also guarding against a statistical trap: the population of approved borrowers is shaped by previous policy, so observed performance is not a neutral sample of everyone who might have applied.

The payment loop: instruction is not the same as final settlement

A modern payment can be decomposed into initiation, authentication, authorisation, message routing, clearing, settlement, receiver credit, reconciliation and exception handling. Different systems compress or rearrange these stages, but the distinctions matter. The customer interface may report success before every back-office process has finished. A payment message can create an obligation between institutions before the obligation is finally discharged in settlement money.

If payer and payee use the same bank, the transaction may be resolved largely through internal ledger changes: one deposit liability falls and another rises. If they use different banks, an interbank obligation appears. The banks need a rule for how to calculate what each owes and a settlement asset in which to discharge the obligation. Central-bank money is commonly used for final settlement in systemically important interbank arrangements because it removes credit exposure to a private settlement bank at that final layer.

This distinction explains why payment-system mathematics includes graphs, queues, netting, liquidity-saving mechanisms, cut-off times and finality rules. The visible retail payment is one edge of a much larger network. A systems article should trace both the customer balance and the interbank balance. Otherwise it stops at the interface and misses the financial infrastructure that makes the promise real.

Clearing: work out the obligations before or during settlement

Clearing is the set of processes that transmit, reconcile and, in some systems, net obligations before final settlement. A simple bilateral example shows the effect of netting. If Bank A owes Bank B 100 and Bank B owes Bank A 80, gross obligations total 180 but the net obligation is 20 from A to B. Netting can reduce the amount of settlement liquidity required, but it also changes the system’s dependency on rules, timing and the successful completion of a cycle.

The benefit is not free. Netting creates an interval during which obligations can accumulate. If a participant fails before settlement, the system needs loss-allocation, collateral, default-management or unwind rules, depending on the infrastructure. Real financial market infrastructures therefore pair efficiency with risk controls. The BIS–IOSCO Principles for Financial Market Infrastructures make credit, liquidity and operational risk management central requirements rather than treating clearing as a clerical detail.

Mathematically, clearing can be represented as a matrix of bilateral obligations. Rows can represent payers, columns receivers, and net positions can be derived by subtracting total incoming obligations from total outgoing obligations. More advanced network models ask whether a default propagates through the matrix, whether collateral absorbs a shortfall, whether a central counterparty changes the exposure graph, and whether liquidity-saving algorithms alter queue dynamics.

Settlement: when the obligation is finally discharged

Settlement is the completion stage at which an obligation is discharged through the transfer of the relevant settlement asset. The World Bank describes clearing and settlement as core payment-system functions, and the Federal Reserve’s payment-system-risk policy emphasises the credit, liquidity, operational and legal risks that surround these processes. In securities markets, settlement may require delivery-versus-payment; in foreign exchange, payment-versus-payment can reduce principal risk.

The timing of settlement changes risk. Gross real-time settlement can reduce the time an unpaid obligation remains outstanding, but it can require more intraday liquidity because each payment may need to settle individually. Deferred net settlement can economise on liquidity by offsetting obligations, but exposures can accumulate until the settlement cycle completes. There is no universal “fastest is always safest” rule; modern research shows trade-offs among liquidity cost, counterparty risk, network structure and settlement timing.

A closed-loop model follows what settlement consumes and what it releases. Settling a payment uses reserves or another settlement asset. Receipt of settlement replenishes the receiving institution. Queued payments can create delays; incoming funds can unlock outgoing payments. The system can therefore exhibit liquidity recycling: the same unit of settlement asset supports multiple payments sequentially during the day. Throughput depends on timing as well as quantity.

Liquidity, capital and solvency are different constraints

Liquidity answers: can the bank meet obligations when they come due? Capital answers: how much loss-absorbing residual funding stands behind the bank’s assets? Solvency asks whether the value of assets exceeds liabilities. The Federal Reserve explicitly distinguishes liquidity from capital: liquid assets can meet near-term obligations, while capital absorbs losses. A bank can be solvent on a valuation basis yet face a liquidity crisis if it cannot turn assets into settlement-ready funds fast enough. A bank can be liquid today yet economically insolvent if asset losses exceed equity.

This creates a two-dimensional survival problem. One axis is value; the other is time. An asset may be valuable in the long run but difficult to sell quickly without a discount. A sudden deposit outflow accelerates the time constraint. If the bank sells assets at depressed prices to raise liquidity, realised losses can erode capital. Falling capital can damage confidence and raise funding costs, which can produce more outflows. A liquidity problem can therefore become a solvency problem through a feedback loop.

The reverse can also happen. A bank with weak capital may face higher funding costs or reduced market access because counterparties fear loss. That worsens liquidity. The correct model therefore resists slogans such as “capital is cash” or “liquidity solves insolvency.” They are coupled but distinct state variables. Robust systems monitor both, plus the speed at which one can deteriorate into the other.

Bank capital: the loss-absorbing loop

Equity absorbs losses before ordinary deposit liabilities are reduced in normal going-concern accounting. If a bank earns profit and retains it, equity can grow. If loans default or securities lose value in a way that reduces recognised assets, equity can fall. Capital regulation then adds definitions and ratios that determine what qualifies as regulatory capital and how exposures are measured. The closed-loop idea is simple even when the rules are complex: risk and losses change capital; capital changes feasible balance-sheet action; new action changes future risk and earnings.

Suppose a bank has 100 of qualifying capital against 1,000 of a simplified risk measure, giving a 10% ratio. If losses reduce qualifying capital to 90 while the denominator is unchanged, the ratio falls to 9%. If the bank wants to restore 10%, it can rebuild the numerator, reduce or alter the denominator, or combine both. Each route has different economic consequences. Retaining earnings takes time; issuing equity changes ownership; shrinking assets can reduce credit supply; changing portfolio composition can shift risk elsewhere.

A systems reader should also distinguish accounting equity, regulatory capital and market capitalisation. They are related but not identical. Regulatory frameworks apply eligibility rules and deductions; market value reflects investor expectations. Mixing the concepts can produce false conclusions. Closed-loop reasoning works only if each variable has a precise definition.

Liquidity: the timing and buffer loop

Liquidity is about the ability to fund assets and meet obligations as they come due. Banks hold liquid assets, maintain funding sources, manage maturity profiles and monitor cash-flow timing. Payment flows create intraday liquidity needs; deposit withdrawals create funding outflows; collateral can be mobilised; central-bank facilities may provide backstop liquidity subject to rules and eligibility. The key mathematical variables are amount, timing, uncertainty, convertibility and haircut.

A simple liquidity ladder groups expected inflows and outflows by time bucket. The bank can compute cumulative net cash flow and compare it with available liquidity. Stress testing then changes assumptions: deposits run off faster, wholesale funding does not roll, collateral values fall, margin calls rise, or contingent facilities are drawn. The output is not a prophecy. It is a conditional statement about how long the institution might meet obligations under a specified scenario.

Time is the critical nonlinearity. If outflows accelerate, management has less time to sell assets, raise funds or change behaviour. Digital communication and instant payments can compress decision times. At the same time, real-time monitoring and automated collateral processes can improve response. Technology therefore changes both the disturbance and the controller.

Interest rates: a system-wide feedback variable

Interest rates connect central-bank policy, wholesale funding, deposit pricing, loan demand, borrower affordability, asset valuation and bank profitability. A policy-rate change does not pass through mechanically at one speed. Deposit rates may adjust differently from mortgage or corporate-loan rates; fixed-rate assets reprice later than floating-rate liabilities; securities prices move immediately; customer behaviour changes with delays. The banking system therefore contains multiple transmission lags.

A rise in rates can improve income on some assets but raise funding cost, lower the market value of fixed-rate securities, reduce borrowing demand and increase debt-service pressure. The sign of the net effect depends on repricing gaps, hedges, customer behaviour, asset mix and time horizon. This is a classic reason to model the balance sheet as a dynamic system rather than treat “higher rates” as one-directional.

The feedback loop can be written qualitatively: policy rate → market rates → deposit and funding rates → loan pricing and asset values → borrower behaviour and bank income → credit quality and capital → lending standards and credit growth → economic activity and inflation → future policy. Each arrow has uncertainty, delay and institution-specific sensitivity.

Profitability: the replenishment loop

Banks need earnings not merely as a commercial objective but because retained profit can rebuild capital and fund investment in systems, staff and resilience. A simplified income loop starts with interest income and fee income, subtracts funding cost, operating cost and credit losses, then passes through tax and distributions to retained earnings. Retained earnings can increase equity. Equity supports future risk-taking capacity subject to constraints.

Profitability can also create dangerous positive feedback if incentives reward volume without adequate attention to risk. Rapid asset growth generates revenue today while credit losses arrive later. If the performance window is shorter than the risk horizon, a system can appear successful while accumulating latent failure. Closed-loop governance therefore asks whether the measurement period captures the return of consequences.

A high-quality system uses lagged outcomes. Underwriting teams should eventually see defaults and recoveries, not only origination volume. Pricing teams should see full-life profitability, not just initial margin. Treasury should see funding and liquidity consequences. Operations should see exception and fraud losses. Closing the measurement loop reduces the chance that one department optimises a local metric by exporting risk to another.

Risk is not one variable

Banking risk is a vector, not a scalar. Credit risk concerns borrower or counterparty failure. Market risk concerns adverse price moves. Liquidity risk concerns inability to meet obligations on time. Operational risk includes process, systems, human and external-event failures. Legal risk concerns enforceability and rules. Model risk concerns decisions based on wrong, unstable or misused models. Cyber risk can become operational and liquidity stress. Strategic and reputational effects can change behaviour and funding.

The risks interact. A market shock can create margin calls, producing liquidity stress. Liquidity stress can force asset sales, deepening market losses. Market losses can reduce capital. Weak capital can raise funding cost and trigger confidence effects. Operational outages can delay payments and reconciliation, creating liquidity uncertainty. A fraud event can cause direct loss and secondary reputational outflows. Systems analysis looks for these cross-risk transitions rather than maintaining isolated risk boxes.

The mathematical task is to map conditional dependencies. Correlation is not enough. Timing and direction matter. A directed graph can represent one risk causing another. Scenario analysis can activate several edges at once. Stress testing can ask which combination breaches a constraint. Reverse stress testing begins at failure and searches backward for plausible paths. The goal is not to produce a single magic probability; it is to understand pathways.

Collateral, margin and haircut loops

Collateral turns an unsecured exposure into a claim supported by pledged assets, but it introduces valuation and liquidity feedback. If collateral value falls, a lender or clearing system may require more collateral through a margin call or haircut adjustment. The borrower must find cash or eligible assets. That demand can force sales. If many participants sell similar assets at once, prices can fall further, causing more margin calls—a classic procyclical loop.

A haircut is a risk adjustment between market value and lending value. If an asset worth 100 receives a 20% haircut, it may support only 80 of secured funding. Raising the haircut to 30% reduces funding capacity to 70 without changing the quantity of the asset. Haircut changes therefore transform market-value movements into liquidity constraints. The direction of feedback depends on design: conservative buffers can improve resilience, while rapidly varying requirements can amplify stress.

Central counterparties and secured funding markets manage these risks with margin models, default resources, eligibility rules and stress tests. The specialist mathematics can be complicated, but the loop is intuitive: position → exposure → collateral requirement → funding need → market action → price change → new exposure. The system closes when the updated market state returns to the next margin calculation.

Financial market infrastructure: the hidden coordination layer

Financial market infrastructures include payment systems, central counterparties, securities settlement systems, central securities depositories and trade repositories. The Reserve Bank of Australia describes them as key components of the financial system that support netting, clearing, settlement and default coordination. The Bank of England likewise treats FMIs as vital to financial stability because they allow transactions to be cleared, settled and recorded at scale.

An FMI is not just plumbing in the sense of being passive. Its rules shape participant incentives, liquidity demand, exposure concentration and recovery behaviour. A central counterparty can reduce bilateral exposure complexity by interposing itself between buyers and sellers, yet it also concentrates risk-management responsibilities in the CCP. A real-time gross settlement system can reduce settlement exposure while increasing intraday liquidity needs. A queue rule can determine which payments settle first under scarcity.

Systems mathematics asks whether the infrastructure has enough redundancy, liquidity, operational capacity and recovery tools to preserve critical functions under stress. Network topology matters: a highly central node can improve efficiency while becoming systemically important. The right design question is not “centralised or decentralised?” in the abstract; it is what failure modes each architecture creates, contains or shifts.

Central banks: settlement asset, policy transmitter and backstop

Central banks sit at several loops simultaneously. They provide central-bank money used for settlement, implement monetary policy, oversee or operate parts of payment infrastructure in many jurisdictions, and may provide liquidity to eligible institutions under defined conditions. Their role makes the banking system hierarchical: customers hold commercial-bank deposits; banks settle among themselves using central-bank money.

This hierarchy helps explain why a transfer between two customers at different banks ultimately involves a claim between banks that can be discharged in central-bank reserves. It also explains why central-bank liquidity can matter during stress. A solvent institution facing temporary liquidity pressure may have assets but insufficient immediately available settlement money. A collateralised central-bank facility can transform eligible collateral into liquidity, buying time. It cannot make underlying economic losses disappear.

Monetary policy creates another feedback path. The central bank observes inflation, activity, financial conditions and other information, changes policy settings, and waits for transmission through market rates, bank pricing, borrowing, saving and spending. The economy then returns new data. This is a feedback system with long and uncertain delays, which is why policy control is harder than a simple thermostat analogy suggests.

Supervision and regulation: constraints are part of the state machine

Banking does not operate under purely internal optimisation. Laws, capital and liquidity requirements, large-exposure limits, governance expectations, consumer-protection rules, anti-money-laundering obligations, payment-system rules, recovery planning and resolution frameworks shape feasible actions. In a mathematical model, these are constraints and state-transition rules. They change what the institution can do after a shock.

A regulatory ratio is not merely a reporting statistic. If a limit becomes binding, management behaviour may change: raise capital, alter assets, increase liquidity, reduce concentration, change funding or suspend distributions. That reaction feeds back into markets and the real economy. If many institutions react similarly at the same time, individually prudent behaviour can produce system-wide effects, such as asset sales or credit contraction. Macroprudential policy exists partly because institution-level optimisation does not automatically produce system-level stability.

Good analysis therefore distinguishes microprudential and system-wide objectives. A rule can strengthen one institution while shifting risk to non-banks; a requirement can reduce one failure mode while increasing another cost. The relevant question is the whole response path, including adaptation.

Resolution: the loop after ordinary controls fail

A robust system plans not only for prevention but for failure. Resolution frameworks aim to preserve critical functions, allocate losses according to legal rules and avoid uncontrolled contagion when a bank can no longer continue normally. Recovery is what the institution does while still viable; resolution is an authority-led process for handling failure. Deposit insurance can reduce incentives for insured depositors to run, but coverage, eligibility and institutional arrangements differ by jurisdiction.

The closed-loop reason for resolution planning is that failure itself produces new obligations and feedback. Payments must continue, deposits may need transfer or payout, counterparties revalue exposures, collateral moves, markets react, and surviving institutions absorb flows. A disorderly closure can turn a local loss into network-wide uncertainty. A credible resolution path changes behaviour before failure by influencing expectations about who bears loss and whether critical services continue.

For mathematics students, resolution resembles a state transition under boundary breach. Normal operating equations no longer apply unchanged. The system moves into a different rule set with different priorities and control authority. Modelling failure therefore requires regime switches, not just extending normal-state averages.

The real-economy world-return loop

Finance is useful because it moves purchasing power, risk and claims across time. A business borrows to buy equipment, hire staff or finance inventory. A household borrows to bring forward a purchase and repays from future income. Investors fund enterprises in exchange for claims on future value. Governments issue debt to finance public activity. The financial system sends resources outward; the real economy returns wages, revenues, taxes, rents, profits, defaults and asset values.

This return path is the ultimate test of many financial claims. The bank’s model can estimate probability of default, but the borrower’s future cash flow decides whether contractual payments occur. A security can be liquid in ordinary markets, but stress can change buyers’ willingness. A collateral valuation can be precise at noon and obsolete after a market gap. Models are maps of possible return paths, not substitutes for the world.

The strongest educational model therefore keeps two ledgers: the financial ledger of claims and the real ledger of resources and activity. Loan creation changes the financial ledger immediately. The funded factory, home, education, inventory or working capital changes the real world over time. Confusing the two makes finance appear self-contained. Connecting them explains why credit quality, productivity, income and asset values matter.

Feedback signs: stabilising and amplifying loops

A negative-feedback loop counteracts deviation. If liquidity falls, a bank can reduce new lending, raise funding, sell liquid assets or mobilise collateral, tending to restore the buffer. If capital falls, retained earnings or new equity can rebuild it. A positive-feedback loop amplifies movement. Falling asset prices can trigger margin calls, forced sales and further price declines. Deposit outflows can cause asset sales, losses and more concern.

The labels “positive” and “negative” do not mean good and bad. They describe the sign of feedback. Positive feedback reinforces change; negative feedback offsets it. A well-designed system often needs both. Credit expansion during recovery can support activity, but unchecked reinforcing growth can create leverage. Emergency liquidity can stabilise a run, but poorly designed guarantees can alter incentives. Systems engineering focuses on magnitude, delay, saturation and side effects.

The same loop can change sign across regimes. Higher interest rates may improve margins when assets reprice faster than liabilities, then reduce margins later when deposits reprice or credit losses rise. Asset sales may restore liquidity for one bank but depress market prices for all banks if many sell together. Context determines the effective feedback coefficient.

Delays: the hidden variable that turns control into instability

Every major banking loop has delay. Credit decisions occur today; defaults may appear years later. Policy rates change today; deposit behaviour and investment respond over months. A fraud alert appears in milliseconds; investigation and recovery can take much longer. A payment instruction travels quickly; reconciliation or dispute resolution may continue after settlement. Delays matter because a controller acts on information about a system that may already have moved.

In control theory, excessive delay can make a stabilising response overshoot. Banking has analogous problems. If loan growth is judged by recent arrears, a rapidly growing portfolio may look excellent simply because risky loans have not seasoned long enough to fail. By the time defaults reveal the true state, exposure may be much larger. This is why vintage analysis and forward-looking indicators matter.

Delay also affects liquidity. A payment queued for ten minutes may be harmless in one system and critical in another if downstream obligations depend on it. Intraday liquidity models therefore care about timestamps, not only end-of-day totals. An institution that always finishes the day balanced can still experience dangerous intraday peaks.

Networks: local safety does not guarantee system safety

Banks, payment providers, FMIs, markets and customers form networks. A bank can have bilateral exposures to many institutions, common asset holdings with others, shared service providers and common funding markets. Network risk arises from direct links and from indirect similarity. Two banks that never lend to each other can still transmit stress by selling the same securities into the same thin market.

Graph mathematics provides a useful language. Nodes represent institutions or accounts; edges represent payments, exposures or dependencies. Edge weights can encode amount, probability, capacity or time. Degree measures direct connectivity. Betweenness can identify nodes lying on many paths. Centrality measures can reveal concentration, but no metric alone proves systemic importance because substitutability, legal structure, liquidity and operational function also matter.

The existing BTT article How Interbank Networks Transmit Bank Stress owns that specialist mechanism. In this master article, the network is one layer in the larger loop: exposures create losses; losses change capital and liquidity; responses change market prices and payment behaviour; those changes return through the network.

The mathematical toolkit for closed-loop banking and finance

No single equation captures the system. Different mathematical tools answer different questions. Accounting identities preserve balance-sheet consistency. Difference equations track stocks and flows. Probability models estimate uncertain events. Statistics calibrate and validate models. Graph theory represents payment and exposure networks. Queueing theory studies payment congestion. Optimisation allocates scarce liquidity or capital. Control theory describes feedback. Game theory studies strategic behaviour. Simulation explores nonlinear paths.

A practical modelling sequence is: define the system boundary; list state variables; classify stocks and flows; write identities; define transition rules; map delays; identify constraints; specify feedback; choose stochastic elements; test edge cases; compare model output with observed data; and state what would falsify the model. The sequence is more important than mathematical decoration. A simple model with explicit boundaries is more useful than a complicated model whose variables are ambiguous.

The final step is interpretation. A 99th-percentile loss, a liquidity survival horizon, a capital ratio, a centrality score and a payment queue length are different objects. They cannot be averaged into one “safety score” without a defensible model. Good mathematics preserves meaning.

Singapore: from PayNow and FAST to final settlement

Singapore provides a compact real-world illustration because everyday retail payments connect to national infrastructure. The Association of Banks in Singapore states that PayNow lets participating bank and major-payment-institution customers send Singapore-dollar funds using identifiers such as mobile numbers or UENs and that the transfer runs through FAST. The same site notes current changes in the e-payment landscape, including Electronic Deferred Payment and EDP+ introduced in 2025 as alternatives for deferred payments.

At the infrastructure layer, the Monetary Authority of Singapore’s Financial Institutions Directory currently lists FAST, Inter-bank GIRO, the Singapore Dollar Cheque Clearing System, NETS EFTPOS and the New MAS Electronic Payment and Book-Entry System among designated payment-system activities, with MAS acting as settlement institution for several systems. The important educational point is the hierarchy: a convenient retail alias such as a mobile number is not itself the settlement asset. The user interface, payment rail, clearing process and settlement layer are distinct.

A Bukit Timah family sending a PayNow transfer therefore provides a useful systems story without requiring any fictional claim about a specific bank. Alicia can enter a recipient identifier; her bank authenticates and authorises the instruction; the payment traverses participating infrastructure; obligations between institutions are cleared and settled according to the system’s rules; the receiver’s account is credited; both institutions reconcile records. What looks like one tap is a coordinated state transition across several ledgers.

Current local facts change, so the article should not freeze operational details as timeless truths. The appropriate authoritative references are MAS’s designated payment-system directory and the Association of Banks in Singapore’s PayNow page. The mathematical architecture remains useful even as products and operators evolve.

Alicia, Tricia and Kai Kai: three ways to read the same system

Alicia starts with the customer-visible event. She asks: “I pressed send. What changed?” Her ledger balance changes, but she learns to trace authentication, message routing, clearing, settlement and reconciliation. Her habit is to refuse the interface as the endpoint. Every time she sees a transaction, she asks for the hidden state changes beneath it.

Tricia starts with the balance sheet. She asks: “Which side is this on?” A customer deposit is the bank’s liability; a loan is the bank’s asset; reserves are assets; equity absorbs losses. Her habit is to preserve the accounting identity before reasoning about policy. When someone says a bank “uses deposits to lend”, she asks whether the statement is about funding, liquidity, money creation, or accounting—because those are related but different questions.

Kai Kai starts with failure. He asks: “What has to be true for this loop to continue?” A payment system needs operational availability and settlement liquidity. A loan needs future borrower cash flow. A bank needs both liquidity and solvency. A model needs calibrated assumptions that remain relevant. His habit is to search for the return condition and the failure boundary.

Together the trio makes a robust systems method: Alicia follows the transaction, Tricia reconciles the state, and Kai Kai tests the boundary. The method works far beyond banking. Any closed-loop system can be examined by tracking events, state variables and failure conditions.

How loops break

Closed loops fail in several recurring ways. The return is delayed beyond the system’s tolerance. The returning signal is wrong. The feedback has the wrong sign. A buffer is too small. The model observes the wrong variable. An external shock exceeds design assumptions. A network dependency has no substitute. Legal finality is unclear. Operational systems are unavailable. Participants adapt to the rule and change the behaviour the rule was designed around.

In banking, these abstract failures have concrete forms. A borrower’s cash flow arrives too late to meet debt service. A bank’s liquid assets cannot be monetised quickly enough. Margin calls arrive before collateral can be mobilised. A fraud model creates too many false positives and operations become overloaded. A stress model underestimates correlated outflows. A payment outage creates reconciliation uncertainty. A capital buffer absorbs losses but market confidence disappears faster than capital can be rebuilt.

The correct response is not to claim a system is “safe” because one control exists. Robustness is layered. Capital, liquidity, collateral, operational redundancy, governance, supervision, resolution and credible information each address different failure modes. Diversity of controls is valuable precisely because the future disturbance is uncertain.

What would falsify a closed-loop explanation?

A useful model must expose conditions under which it would be wrong. If a model says faster settlement always reduces risk, evidence of liquidity-driven failure under faster settlement challenges the claim. If a model says deposits are stable, rapid observed outflows falsify the behavioural assumption. If a credit model predicts calibrated 2% default probability for a cohort and comparable cohorts repeatedly default at 8%, the calibration is wrong even if the ranking remains useful.

Falsification is not humiliation; it is system maintenance. Models should have update triggers: regime change, material forecast error, new regulation, new product design, data drift, infrastructure change, crisis behaviour or structural shift in customer response. A closed-loop model should itself be closed loop: prediction → observation → error → diagnosis → recalibration → new prediction.

This epistemic loop is one of the most important ideas in technical finance. The system changes partly because participants learn. A fixed model operating in an adaptive world accumulates error. Versioning, backtesting and challenger models are therefore mathematical necessities, not administrative decoration.

A closed-loop design checklist

• Define the boundary: which institution, market, infrastructure and real-economy interactions are inside the model?
• Name the state variables: deposits, reserves, loans, capital, collateral, queued payments, exposures, prices, funding and other stocks.
• Name the flows: originations, repayments, withdrawals, settlements, fees, losses, margin calls and funding flows.
• Separate message, obligation and settlement.
• Separate liquidity from capital and both from solvency.
• Map time: intraday, overnight, monthly, contractual maturity and crisis horizons.
• Map feedback signs and delays.
• Identify constraints: legal, operational, capital, liquidity, collateral and risk limits.
• Model network dependencies and common exposures.
• Include the real-world return path.
• State failure conditions and recovery states.
• Define observations that would falsify or force an update to the model.

Closed-loop diagnostic atlas: 50 systems, one method

1. Deposit funding

Reader job: understand how customer balances support funding while remaining payable liabilities. The relevant state is deposit liabilities and liquid assets. Inflows include incoming transfers, salary credits, new account balances; outflows include payments, withdrawals, transfers and principal repayments. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: pricing and confidence change future inflows and outflows. Failure boundary: concentrated or rate-sensitive deposits can leave faster than replacement funding arrives. A practical quantitative dashboard can therefore include deposit beta, decay rate, concentration, runoff and survival horizon. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

2. Loan origination

Reader job: trace how credit approval creates a claim and changes the balance sheet. The relevant state is loan assets and associated deposit liabilities. Inflows include approved principal and drawdowns; outflows include repayment, sale, write-off or maturity. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: realised performance updates underwriting and pricing. Failure boundary: growth can outrun observed defaults because losses arrive with delay. A practical quantitative dashboard can therefore include probability of default, expected loss, vintage curves and approval rates. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

3. Mortgage repayment

Reader job: connect amortisation to household cash flow and bank asset runoff. The relevant state is outstanding principal. Inflows include capitalised interest where applicable and additional advances; outflows include scheduled principal, prepayment and default resolution. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: rates and refinancing incentives alter prepayment. Failure boundary: prepayment or default can diverge sharply from static schedules. A practical quantitative dashboard can therefore include recurrence relations, hazard rates and duration. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

4. Credit-card revolving balances

Reader job: model a line that can be repeatedly drawn and repaid. The relevant state is utilised credit and available line. Inflows include purchases, cash advances and fees; outflows include payments, charge-offs and account closure. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: utilisation and delinquency update line management. Failure boundary: behaviour changes rapidly near financial stress. A practical quantitative dashboard can therefore include utilisation rate, roll rates, exposure at default and cure. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

5. Interbank payment

Reader job: follow an obligation between banks from message to settlement. The relevant state is customer deposits, interbank positions and reserves. Inflows include incoming payment obligations and received settlement; outflows include outgoing payment obligations and reserve transfers. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: settlement receipts replenish capacity for later payments. Failure boundary: queues and liquidity scarcity can delay otherwise valid payments. A practical quantitative dashboard can therefore include gross flow, net position, queue time and intraday liquidity. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

6. FAST retail transfer

Reader job: separate customer convenience from infrastructure state. The relevant state is account balances and bank settlement positions. Inflows include authorised incoming transfers; outflows include authorised outgoing transfers and settlement. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: successful processing reinforces trust and usage. Failure boundary: operational or liquidity disruption can create exceptions. A practical quantitative dashboard can therefore include availability, latency, settlement position and reconciliation breaks. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

7. PayNow alias resolution

Reader job: distinguish addressing from money movement. The relevant state is registered aliases and linked destination accounts. Inflows include new registrations and updates; outflows include deregistration or reassignment. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: verification quality changes error and fraud rates. Failure boundary: an alias can be correct syntactically but wrong economically if user validation fails. A practical quantitative dashboard can therefore include resolution success, confirmation errors and exception rates. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

8. Card authorisation

Reader job: separate approval from clearing and settlement. The relevant state is available balance or credit line and pending authorisations. Inflows include new approved transactions; outflows include reversals, clearing presentments and settlement. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: fraud and dispute outcomes update decision rules. Failure boundary: authorisation success can still end in later exception or chargeback. A practical quantitative dashboard can therefore include approval rate, false-positive rate, fraud loss and latency. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

9. Clearing netting

Reader job: show how offsetting obligations change settlement liquidity. The relevant state is gross bilateral obligations and net positions. Inflows include new payment or trade obligations; outflows include netting and settlement completion. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: participant failure can change expected net positions. Failure boundary: efficiency gains create dependency on cycle completion and rules. A practical quantitative dashboard can therefore include gross-to-net ratio, largest net debit and concentration. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

10. Real-time gross settlement

Reader job: study payment finality under high-frequency liquidity demand. The relevant state is reserve balances and queued obligations. Inflows include incoming settled payments and liquidity injections; outflows include outgoing settled payments. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: incoming liquidity unlocks queued outgoing transactions. Failure boundary: insufficient intraday liquidity can create gridlock. A practical quantitative dashboard can therefore include throughput, queue length, turnover and peak liquidity. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

11. Bank reserves

Reader job: track the settlement asset at the central bank. The relevant state is reserve account balance. Inflows include incoming settlement, central-bank operations and eligible funding; outflows include outgoing settlement and reserve drains. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: price and availability influence bank liquidity decisions. Failure boundary: aggregate and individual reserve positions can behave differently. A practical quantitative dashboard can therefore include reserve balance, intraday low point and marginal liquidity cost. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

12. Liquidity buffer

Reader job: measure capacity to survive stressed cash outflow. The relevant state is high-quality liquid assets and immediately available funding. Inflows include maturing inflows and new funding; outflows include withdrawals, margin calls and maturing liabilities. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: stress experience updates buffer targets. Failure boundary: assets may lose liquidity exactly when they are needed. A practical quantitative dashboard can therefore include coverage ratio, cash-flow gap and survival horizon. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

13. Wholesale funding

Reader job: model dependence on market refinancing. The relevant state is outstanding wholesale liabilities. Inflows include new issuance, repo and interbank borrowing; outflows include maturity, withdrawal and non-rollover. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: market confidence changes spreads and access. Failure boundary: refinancing can fail discontinuously rather than gradually. A practical quantitative dashboard can therefore include maturity ladder, rollover probability, spread and concentration. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

14. Bank capital

Reader job: trace how profit and loss change loss-absorbing capacity. The relevant state is qualifying capital and equity. Inflows include retained earnings and new issuance; outflows include losses, distributions and deductions. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: capital ratios constrain future asset growth. Failure boundary: risk-weight or valuation changes can alter ratios without cash movement. A practical quantitative dashboard can therefore include capital ratio, leverage ratio, retained earnings and buffer. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

15. Credit provisions

Reader job: connect expected loss estimates to earnings and capital. The relevant state is allowances and loan carrying values. Inflows include new expected-loss recognition; outflows include write-offs, recoveries and releases. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: realised outcomes recalibrate expected loss. Failure boundary: model optimism can delay recognition of deterioration. A practical quantitative dashboard can therefore include stage migration, coverage, loss rate and forecast error. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

16. Interest-rate repricing

Reader job: follow policy and market rates into bank cash flows. The relevant state is fixed and floating rate assets and liabilities. Inflows include repricing events and new business; outflows include maturity and refinancing. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: customer response changes volumes and spreads. Failure boundary: different repricing speeds create temporary gains or losses. A practical quantitative dashboard can therefore include duration gap, beta, margin and key-rate sensitivity. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

17. Net interest margin

Reader job: connect asset yield and funding cost to retained earnings. The relevant state is interest-earning assets and interest-bearing liabilities. Inflows include interest income; outflows include interest expense and credit costs. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: profit changes capital and pricing strategy. Failure boundary: headline margin can improve while future credit risk worsens. A practical quantitative dashboard can therefore include yield, funding cost, NIM and risk-adjusted margin. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

18. Collateral valuation

Reader job: translate market prices into secured funding capacity. The relevant state is eligible collateral inventory. Inflows include new collateral and appreciation; outflows include release, haircut increase and depreciation. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: price changes trigger margin and funding actions. Failure boundary: procyclical haircuts can amplify market stress. A practical quantitative dashboard can therefore include market value, haircut, lendable value and concentration. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

19. Margin calls

Reader job: follow exposure changes into liquidity demand. The relevant state is posted collateral and available liquidity. Inflows include collateral receipts; outflows include variation and initial margin payments. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: large calls force funding or position reduction. Failure boundary: simultaneous calls across firms can create fire-sale pressure. A practical quantitative dashboard can therefore include margin-at-risk, call size, timing and liquidity coverage. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

20. Central counterparty

Reader job: understand novation and concentrated default management. The relevant state is member positions, margin and default resources. Inflows include new cleared trades and collateral; outflows include settlement, close-out and loss allocation. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: member risk changes margin and default-fund needs. Failure boundary: risk concentration in the CCP raises resilience requirements. A practical quantitative dashboard can therefore include initial margin, default fund, stress loss and concentration. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

21. Securities settlement

Reader job: join transfer of ownership with transfer of cash. The relevant state is securities positions and cash balances. Inflows include purchases and receipts; outflows include deliveries and cash payments. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: failed settlement changes future liquidity and inventory. Failure boundary: principal risk arises if one leg completes without the other. A practical quantitative dashboard can therefore include fail rate, DvP completion, settlement duration and liquidity. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

22. Foreign-exchange settlement

Reader job: track two currency legs and time-zone exposure. The relevant state is currency balances and FX obligations. Inflows include received currency; outflows include delivered currency. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: settlement success affects limits and future trading. Failure boundary: one currency leg can be paid while the other fails without PvP protection. A practical quantitative dashboard can therefore include Herstatt exposure, PvP coverage, netting and funding need. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

23. Correspondent banking

Reader job: model cross-border chains through intermediary accounts. The relevant state is nostro/vostro balances and payment obligations. Inflows include incoming cross-border funds; outflows include outgoing transfers, fees and settlement. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: compliance and liquidity experience changes routing. Failure boundary: long chains create opacity, delay and trapped liquidity. A practical quantitative dashboard can therefore include hop count, fee, latency, rejection and prefunding. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

24. Treasury asset-liability management

Reader job: coordinate maturity, rate and liquidity structure. The relevant state is asset and liability maturity buckets. Inflows include new funding and asset cash inflows; outflows include maturities, withdrawals and asset purchases. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: measured gaps change hedging and funding choices. Failure boundary: optimising one metric can worsen another risk dimension. A practical quantitative dashboard can therefore include gap, duration, liquidity ladder and hedge effectiveness. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

25. Funds transfer pricing

Reader job: send internal funding and liquidity costs back to business lines. The relevant state is internal transfer-price curves and business balances. Inflows include new funding credits; outflows include charges for asset funding and liquidity usage. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: business pricing changes origination mix. Failure boundary: bad internal prices can reward volume that destroys group economics. A practical quantitative dashboard can therefore include FTP spread, behavioural maturity and risk-adjusted return. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

26. Operational reconciliation

Reader job: close the ledger after transactions. The relevant state is booked transactions and unmatched items. Inflows include new records from channels and counterparties; outflows include matched items and resolved exceptions. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: exception patterns update controls. Failure boundary: unresolved breaks can hide financial or operational errors. A practical quantitative dashboard can therefore include match rate, age of break, value at risk and queue size. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

27. Fraud detection

Reader job: balance loss prevention against customer friction. The relevant state is transaction history, alerts and investigation queue. Inflows include new suspicious signals; outflows include cleared alerts, confirmed fraud and recovery. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: outcomes retrain rules and thresholds. Failure boundary: too many false positives can overload reviewers and block legitimate payments. A practical quantitative dashboard can therefore include precision, recall, false-positive rate, loss and review capacity. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

28. AML transaction monitoring

Reader job: route unusual activity for review without equating anomaly with guilt. The relevant state is customer profiles, transaction graphs and alert queues. Inflows include new transactions and risk signals; outflows include closed cases and escalations. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: investigation outcomes update scenarios and models. Failure boundary: poor thresholds create either blind spots or unmanageable alert volume. A practical quantitative dashboard can therefore include alert rate, conversion, backlog and typology coverage. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

29. Cyber resilience

Reader job: connect technology availability to payment and liquidity continuity. The relevant state is system capacity, credentials, backups and recovery state. Inflows include normal traffic and restored services; outflows include failed transactions and degraded capacity. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: incident learning changes controls and redundancy. Failure boundary: a technology outage can become a liquidity or confidence event. A practical quantitative dashboard can therefore include availability, recovery time, data integrity and backlog. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

30. Model risk

Reader job: make models themselves part of the feedback system. The relevant state is model versions, parameters and validation findings. Inflows include new data and challenger results; outflows include retired models and corrected estimates. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: forecast error triggers recalibration. Failure boundary: a model can remain internally elegant while the world changes. A practical quantitative dashboard can therefore include calibration error, discrimination, drift and override rate. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

31. Stress testing

Reader job: translate scenarios into conditional balance-sheet paths. The relevant state is starting balance sheet and risk factors. Inflows include scenario-conditioned income and funding; outflows include losses, outflows and management actions. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: results change limits, capital and contingency plans. Failure boundary: results are only as credible as scenario, model and behavioural assumptions. A practical quantitative dashboard can therefore include capital depletion, liquidity horizon and breach point. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

32. Reverse stress testing

Reader job: start at failure and search for pathways. The relevant state is constraints and failure thresholds. Inflows include candidate shocks and combinations; outflows include mitigation and rejected paths. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: discovered weak links change controls. Failure boundary: plausibility can be harder to assess than mathematical optimisation. A practical quantitative dashboard can therefore include distance to breach, scenario plausibility and pathway count. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

33. Deposit insurance

Reader job: alter depositor incentives and failure handling. The relevant state is insured deposit base and insurance resources. Inflows include premiums and transferred balances; outflows include payouts and resolution transfers. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: credible protection can stabilise behaviour. Failure boundary: coverage limits and operational readiness matter during rapid failure. A practical quantitative dashboard can therefore include insured share, payout readiness and funding capacity. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

34. Bank resolution

Reader job: move a failing institution into a different rule regime. The relevant state is critical functions, liabilities and loss-absorbing resources. Inflows include resolution funding and transferred assets; outflows include loss allocation and asset disposal. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: credible resolution affects pre-failure expectations. Failure boundary: legal and operational complexity can delay execution. A practical quantitative dashboard can therefore include continuity, loss allocation, liquidity and time-to-transfer. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

35. Central-bank liquidity

Reader job: turn eligible collateral into settlement-ready funding under rules. The relevant state is eligible collateral and reserve accounts. Inflows include central-bank credit and reserve provision; outflows include repayment and collateral release. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: facility terms affect preparedness and stigma. Failure boundary: liquidity support cannot erase underlying insolvency. A practical quantitative dashboard can therefore include collateral value, haircut, draw, maturity and repayment. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

36. Monetary-policy transmission

Reader job: follow policy settings through financial conditions. The relevant state is policy rate, reserves and financial contracts. Inflows include new policy decisions and market repricing; outflows include maturing contracts and behavioural adjustment. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: economic data returns to future policy. Failure boundary: long variable lags and changing structures complicate control. A practical quantitative dashboard can therefore include pass-through, lag, credit growth, inflation and activity. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

37. Bank-run dynamics

Reader job: model confidence, withdrawals and forced responses. The relevant state is deposit base, liquid assets and confidence state. Inflows include liquidity support and incoming funds; outflows include withdrawals and collateral calls. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: outflows can create losses that generate more outflows. Failure boundary: digital speed can compress reaction time dramatically. A practical quantitative dashboard can therefore include runoff velocity, uninsured concentration, asset-sale loss and survival time. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

38. Fire-sale feedback

Reader job: connect forced asset sales to market prices and other balance sheets. The relevant state is saleable assets and market depth. Inflows include new buyers and liquidity support; outflows include forced sales. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: price declines trigger valuation losses and further sales. Failure boundary: common holdings turn local stress into system stress. A practical quantitative dashboard can therefore include market depth, price impact, leverage and overlap. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

39. Interbank contagion

Reader job: trace direct exposures and indirect market channels. The relevant state is bilateral claims and common assets. Inflows include payments and recoveries; outflows include defaults and write-downs. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: losses at one node change others’ capital and behaviour. Failure boundary: network structure can create nonlinear cascades. A practical quantitative dashboard can therefore include exposure matrix, centrality, default cascade and loss amplification. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

40. Market liquidity

Reader job: separate ability to trade from funding liquidity. The relevant state is order book depth and dealer balance-sheet capacity. Inflows include new limit orders and dealer capital; outflows include market orders and inventory consumption. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: volatility changes spreads and participation. Failure boundary: market and funding liquidity can reinforce each other negatively. A practical quantitative dashboard can therefore include spread, depth, price impact and turnover. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

41. Securitisation cash flow

Reader job: follow borrower payments through a contractual waterfall. The relevant state is loan pool balance and tranche claims. Inflows include borrower principal and interest; outflows include fees, tranche payments and losses. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: performance changes triggers and allocation. Failure boundary: waterfalls can shift cash sharply after threshold breaches. A practical quantitative dashboard can therefore include OC ratio, attachment point, delinquency and cash diversion. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

42. Insurance-finance interaction

Reader job: map risk transfer and asset-liability matching. The relevant state is reserves, premiums, investments and claims. Inflows include premium and investment income; outflows include claims and expenses. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: loss experience changes pricing and reserves. Failure boundary: correlated market and claim shocks can hit both sides. A practical quantitative dashboard can therefore include loss ratio, reserve adequacy, duration and liquidity. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

43. Non-bank finance

Reader job: track credit and liquidity outside deposit-taking banks. The relevant state is fund shares, repo, securities and credit claims. Inflows include investor inflows and borrowing; outflows include redemptions, margin and debt maturity. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: redemptions can force sales and affect banks through markets. Failure boundary: bank-only models miss risk migration to funds and platforms. A practical quantitative dashboard can therefore include leverage, redemption, liquidity mismatch and interconnectedness. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

44. Stable-value digital money

Reader job: separate payment convenience from reserve and redemption mechanics. The relevant state is issued tokens or balances and backing assets. Inflows include new issuance; outflows include redemption and transfers. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: confidence in backing affects redemption demand. Failure boundary: maturity or liquidity mismatch can create run-like dynamics. A practical quantitative dashboard can therefore include reserve quality, redemption speed, concentration and settlement. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

45. Closed-loop wallet

Reader job: analyse a restricted payment network without confusing it with the whole banking system. The relevant state is stored value and merchant/user balances. Inflows include top-ups and credits; outflows include purchases and withdrawals where permitted. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: network utility affects adoption and float. Failure boundary: interoperability and redemption constraints define the boundary. A practical quantitative dashboard can therefore include active users, float, redemption, acceptance and transaction velocity. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

46. Open-banking data flow

Reader job: treat consented data access as an information loop rather than money settlement. The relevant state is permissions, credentials and data states. Inflows include new consent and data updates; outflows include revocation and expiry. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: better data can change pricing and product choice. Failure boundary: security, consent quality and data interpretation create risk. A practical quantitative dashboard can therefore include consent rate, API availability, error rate and revocation. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

47. Credit-bureau feedback

Reader job: show how repayment behaviour becomes future underwriting information. The relevant state is credit files and scores. Inflows include new account and performance data; outflows include aged-off or corrected information. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: observed behaviour changes future access and pricing. Failure boundary: selection effects can entrench historical policy patterns. A practical quantitative dashboard can therefore include coverage, error rate, score migration and calibration. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

48. Supervisory reporting

Reader job: return institution data to external oversight. The relevant state is regulatory data sets and risk assessments. Inflows include new reports and examinations; outflows include resolved findings and closed actions. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: findings change governance and controls. Failure boundary: late or poor-quality data weakens the feedback loop. A practical quantitative dashboard can therefore include timeliness, data quality, finding severity and remediation age. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

49. Audit and assurance

Reader job: test whether reported state matches underlying evidence. The relevant state is records, controls and assertions. Inflows include new evidence and tests; outflows include resolved exceptions. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: findings update processes and governance. Failure boundary: sampling and scope can miss fast-changing or hidden failure modes. A practical quantitative dashboard can therefore include exception rate, coverage, control effectiveness and repeat findings. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

50. Real-economy borrower cash flow

Reader job: close the claim with external productive or household income. The relevant state is borrower assets, income capacity and debt. Inflows include sales, wages and other receipts; outflows include operating costs, taxes and debt service. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: performance changes credit decisions. Failure boundary: financial models fail when they ignore the source of repayment. A practical quantitative dashboard can therefore include debt-service coverage, free cash flow, volatility and arrears. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

51. Bank profitability

Reader job: connect operating performance to resilience and future capacity. The relevant state is earning assets, funding and equity. Inflows include interest and fee income; outflows include funding, operating and credit costs. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: retained earnings rebuild capital. Failure boundary: short-term profit targets can encourage hidden long-term risk. A practical quantitative dashboard can therefore include ROA, ROE, NIM, cost-income and risk-adjusted return. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

52. Financial stability

Reader job: treat system resilience as an emergent property rather than one ratio. The relevant state is institution states, market conditions and infrastructure capacity. Inflows include buffers, income and liquidity support; outflows include losses, outflows and failures. The analytical discipline is to reconcile the state before and after the event rather than narrate only the visible transaction.

Feedback: policy and private responses alter future system state. Failure boundary: safe institutions can still interact in destabilising ways. A practical quantitative dashboard can therefore include aggregate leverage, liquidity, contagion, market functioning and critical-service continuity. None of those metrics is a complete verdict by itself. The closed-loop question is whether the metric changes early enough, accurately enough and with enough decision authority to alter the next state before a hard constraint is breached.

Twenty-five worked system questions

The following mini-cases are deliberately simplified. They are not forecasts, regulatory calculations or operating instructions. Their purpose is to show how to convert a banking story into a state-transition problem.

1. Deposit runoff

Setup: A bank starts the morning with 1,000 units of a deposit segment. During the day, 120 leave and 35 arrive.

Closed-loop reading: End balance is 915. The gross outflow rate is 12%, while the net decline is 8.5%. Both measures matter: gross flow describes behavioural intensity; net change describes the stock effect. A system can have very large gross flows with a small net change if inflows offset outflows.

2. Loan and deposit creation

Setup: A bank books a 200 loan and credits the borrower’s deposit by 200.

Closed-loop reading: The simplified balance sheet expands by 200 on both sides: a loan asset rises by 200 and a deposit liability rises by 200. The identity remains balanced. If the borrower sends the 200 to another bank, the deposit can leave while the loan asset remains, creating an interbank settlement and funding implication.

3. Principal repayment

Setup: A borrower repays 20 of principal from a deposit held at the same bank.

Closed-loop reading: Ignoring fees and interest, the bank’s loan asset falls by 20 and its deposit liability falls by 20. The balance sheet contracts. Interest payment is different because it enters income rather than reducing principal one-for-one.

4. Netting

Setup: Bank A owes B 100 while B owes A 70 in the same eligible netting cycle.

Closed-loop reading: The bilateral net obligation is 30 from A to B. Gross obligations total 170, so netting reduces immediate settlement amount in this toy example. The trade-off is that the system depends on the netting rules and successful completion of the cycle.

5. Intraday liquidity turnover

Setup: A bank starts with 40 of reserves, receives 100 of settled payments over the day and sends 110.

Closed-loop reading: Ignoring other items, it ends with 30. Yet total outgoing settlement of 110 exceeded the opening balance because incoming funds replenished liquidity. This is why payment capacity depends on timing and turnover, not only the opening stock.

6. Capital loss

Setup: A bank has assets 1,100, liabilities 1,000 and equity 100. An asset loss of 25 is recognised with liabilities unchanged.

Closed-loop reading: Assets become 1,075 and equity becomes 75. The accounting identity still holds. The 25 loss consumes one quarter of starting equity even though it is only about 2.3% of starting assets, illustrating leverage.

7. Liquidity versus solvency

Setup: A bank owns long-dated assets worth 1,100 against liabilities of 1,000 but has only 10 of immediately available settlement liquidity and faces 50 of withdrawals now.

Closed-loop reading: On these simplified values it can be solvent yet illiquid. It needs to transform assets or obtain funding quickly. If forced sales realise losses large enough, the liquidity response can damage solvency.

8. Haircut

Setup: Collateral has market value 100. A lender applies a 15% haircut.

Closed-loop reading: Lendable value is 85. If the haircut rises to 30%, lendable value falls to 70 even if market price is unchanged. A haircut shock therefore acts like a liquidity shock.

9. Margin call

Setup: A derivatives position generates a 12 variation-margin call due today while the firm has 9 of free cash.

Closed-loop reading: The immediate liquidity gap is 3 before considering other resources. The economic position could be profitable over a longer horizon and still fail today if the firm cannot meet the timing constraint.

10. Rate repricing

Setup: A bank has 500 of floating-rate assets that reprice immediately and 400 of deposits whose pricing adjusts with a delay.

Closed-loop reading: A rate rise can initially increase asset income faster than deposit cost, but the sign may change when deposits reprice or borrowers weaken. A closed-loop model therefore needs timing and behaviour, not only balance amounts.

11. Loan growth delay

Setup: A lender doubles originations this year, while serious defaults usually appear after two years.

Closed-loop reading: Current delinquency can remain low even if underwriting worsened because the new cohort has not seasoned. Vintage analysis closes the delayed feedback loop more honestly than a single portfolio average.

12. Payment queue

Setup: Three outgoing payments of 8, 7 and 6 await settlement while only 10 of liquidity is available.

Closed-loop reading: A simple FIFO rule can settle the 8, leaving 2 and blocking the 7 and 6. A different queue or liquidity-saving algorithm might pair later incoming funds or reorder eligible payments. The example shows why rule design changes throughput.

13. Network concentration

Setup: One service provider connects to 80% of a payment network’s participants.

Closed-loop reading: High connectivity can improve efficiency but creates dependency. A centrality statistic is a warning signal, not proof of fragility; resilience depends on redundancy, substitutability, recovery and capacity.

14. Fire sale

Setup: A bank sells 100 of securities into a thin market and the price impact is 5%.

Closed-loop reading: A rough mark-to-market effect is 5 on that block, but the system effect may be larger if the lower market price revalues similar assets held by others. Common holdings turn one institution’s liquidity action into another institution’s capital problem.

15. Deposit beta

Setup: A policy rate rises 100 basis points and a bank raises a deposit rate by 40 basis points.

Closed-loop reading: A simple deposit beta for that change is 0.40. It is a behavioural summary, not a universal constant. Beta can change by product, customer segment, competition and rate regime.

16. Expected loss

Setup: A simplified portfolio has exposure 1,000, probability of default 2% and loss given default 40%.

Closed-loop reading: A basic expected-loss calculation gives 1,000 × 0.02 × 0.40 = 8. Real accounting and regulatory frameworks are more complex, but the structure shows how exposure, probability and severity multiply.

17. Liquidity survival

Setup: A stressed scenario projects cumulative net cash outflow of 15 per day and an available buffer of 90 with no replenishment.

Closed-loop reading: The toy survival horizon is six days. Real models use nonconstant flows, collateral, contingencies and management actions, but the ratio makes the timing concept visible.

18. Reserve recycling

Setup: A bank receives a 30 settlement at 10:00 and uses the same liquidity to settle a 25 payment at 10:01.

Closed-loop reading: The 30 unit inflow supports later outgoing throughput. End-of-day balances conceal the intraday reuse. This is why time-stamped payment data can matter more than daily totals for liquidity management.

19. Profit retention

Setup: A bank earns 12 after tax and distributes 5.

Closed-loop reading: Ignoring other equity changes, retained earnings add 7 to equity. Profitability therefore feeds capital, but only the retained portion contributes through this route.

20. Capital ratio denominator

Setup: Qualifying capital stays at 100 while a simplified risk-weighted denominator rises from 800 to 1,000.

Closed-loop reading: The ratio falls from 12.5% to 10% even without a nominal capital loss. Balance-sheet growth or risk-weight change can tighten a capital constraint.

21. Foreign-exchange principal risk

Setup: A bank pays one currency before receiving the other currency leg.

Closed-loop reading: During the interval it can lose the full principal if the counterparty fails after receiving the first leg. Payment-versus-payment mechanisms are designed to reduce this timing asymmetry by linking completion of the two legs.

22. Reconciliation backlog

Setup: A system creates 1,000 exceptions per day but operations can resolve only 900.

Closed-loop reading: Backlog grows by 100 per day. After ten unchanged days, it grows by about 1,000. The queue is unstable because arrival rate exceeds service capacity; better detection alone cannot solve the capacity mismatch.

23. False-positive burden

Setup: A fraud model flags 5% of one million daily transactions, while only 0.1% are truly fraudulent.

Closed-loop reading: The alert volume is 50,000. Even a useful model can overwhelm human review if precision is poor. System quality therefore includes downstream capacity and customer friction, not merely statistical recall.

24. Policy transmission lag

Setup: A policy change affects market rates immediately, loan approvals over weeks and household spending over months.

Closed-loop reading: The system contains multiple delays. Observing no immediate spending response does not prove no transmission; reacting too quickly to delayed signals can create policy overshoot.

25. World return

Setup: A business borrows 500, invests in equipment and later generates additional annual operating cash flow of 90 before debt service.

Closed-loop reading: The financial claim can be evaluated only by connecting it to the real activity that generates cash. The loan entry itself does not create the 90; the financed productive process does. Closed-loop finance ends in the world, not in the ledger.

Search-intent map: the questions this master article owns and the questions it sends elsewhere

This page owns broad system intent: banking and finance closed loop systems; how the banking system works as a feedback loop; how deposits, loans, payments, clearing, settlement, liquidity, capital and risk connect; and how financial claims return from the real economy. It does not need to steal every specialist keyword. Good architecture sends narrow intent to the narrow owner.

• For general banking mechanism, see How Banking Works | Deposits, Loans, Payments, Money Creation, Risk and Trust.
• For the broader finance machine, see How Finance Works.
• For financial-system plumbing, see How Financial Systems Work | Money, Credit, Markets and Stability.
• For the BTT mathematics directory, see Finance & Banking Algorithms | Applied Mathematics in Real Financial Systems.
• For payments mathematics, see How Payment Systems Move Money.
• For bank capital, see How Bank Capital Models Turn Risk Into Constraints.
• For liquidity stress, see How Banks Stress-Test Liquidity.
• For reconciliation, see How Banks Reconcile Transactions.

Authoritative reference shelf

The underlying domain facts in this article are anchored to competent public sources. The BIS–IOSCO Principles for Financial Market Infrastructures are a core international reference for payment, clearing and settlement infrastructure. The World Bank’s Payment Systems overview covers national payment systems, real-time gross settlement, fast payments, securities settlement and financial infrastructure. The Federal Reserve’s liquidity-versus-capital explainer gives a clean distinction between those two concepts. The Bank of England’s work on new forms of digital money contains a concise account of commercial-bank deposit creation through lending.

For current Singapore infrastructure, use the Monetary Authority of Singapore Financial Institutions Directory and the Association of Banks in Singapore PayNow page. For global financial-market-infrastructure definitions and risk controls, the Reserve Bank of Australia FMI overview and Bank of England FMI supervision overview are useful. For current research on settlement timing and stability, see the Federal Reserve’s 2025 paper revised in 2026, Settlement Speed and Financial Stability.

Glossary of the closed-loop system

Asset: a resource or claim controlled by an entity with expected economic benefit. Liability: an obligation owed by the entity. Equity: residual interest after liabilities are subtracted from assets. Deposit: a customer asset and bank liability. Loan: a borrower liability and lender asset. Reserve: in this context, a bank asset held at the central bank and used for specified settlement and policy functions. Liquidity: capacity to meet obligations on time. Capital: loss-absorbing financial resources under accounting or regulatory definitions. Solvency: the condition that asset value exceeds liabilities under the relevant valuation basis.

Payment: transfer of funds or the process intended to achieve it. Clearing: transmitting, reconciling and sometimes netting obligations before settlement. Settlement: final discharge of an obligation through transfer of the settlement asset. Finality: the legally defined point at which transfer becomes irrevocable and unconditional under the relevant rules. FMI: financial market infrastructure used to clear, settle or record financial transactions. CCP: central counterparty that interposes itself between buyers and sellers in eligible transactions. CSD: central securities depository. DvP: delivery versus payment. PvP: payment versus payment.

Feedback: when consequences of an action return to influence subsequent action. Positive feedback: reinforces change. Negative feedback: counteracts change. State variable: a quantity describing the current system state. Stock: quantity at a point in time. Flow: quantity over an interval. Buffer: reserve capacity intended to absorb disturbance. Constraint: boundary the system must satisfy. Delay: time between cause, observation and response. Regime: a state in which different rules or relationships apply.

The central proposition

Banking and finance work as a chain of claims only when the chain closes. Deposits must remain usable. Credit must face the world and return as repayment or loss. Payments must clear and settle. Liquidity must arrive before obligations fall due. Capital must absorb losses. Information must return quickly enough to change the next decision. Infrastructure must survive participant failure. When any return path breaks, a local transaction can become a system problem.

That proposition is why the closed-loop view is useful. It does not replace balance-sheet accounting, economics, payment engineering, risk management or regulation. It gives them a common grammar: state → action → transfer → consequence → feedback → updated state. Once readers can see that grammar, banking stops looking like a collection of unrelated products and begins to look like a measurable dynamic system.

For Bukit Timah Tutor, the educational objective is equally specific: show the mathematics without pretending mathematics alone decides policy or proves safety. The system is rich enough for algebra, recurrences, probability, networks, queues, optimisation, simulation and control theory. The discipline is to keep each equation attached to the real object it represents, each assumption visible, and each model answerable to the world it is trying to describe.

Extended systems laboratory: two hundred ways to test whether the loop really closes

Laboratory 1: deposits under a sudden rise in outflows

Start with deposits as a state variable rather than a label. Its role is customer funding and payment claims. Apply a sudden rise in outflows. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include speed, concentration and replacement capacity. The native weakness to watch is runoff speed.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is liquidity buffer. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 2: deposits under a market-price shock

Start with deposits as a state variable rather than a label. Its role is customer funding and payment claims. Apply a market-price shock. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include mark-to-market effects, haircuts and forced actions. The native weakness to watch is runoff speed.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is liquidity buffer. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 3: deposits under an operational outage

Start with deposits as a state variable rather than a label. Its role is customer funding and payment claims. Apply an operational outage. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include backlog, manual fallback and reconciliation integrity. The native weakness to watch is runoff speed.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is liquidity buffer. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 4: deposits under a policy-rate shift

Start with deposits as a state variable rather than a label. Its role is customer funding and payment claims. Apply a policy-rate shift. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include repricing gaps, behavioural response and lag. The native weakness to watch is runoff speed.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is liquidity buffer. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 5: deposits under a confidence shock

Start with deposits as a state variable rather than a label. Its role is customer funding and payment claims. Apply a confidence shock. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include withdrawal behaviour, spreads and market access. The native weakness to watch is runoff speed.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is liquidity buffer. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 6: deposits under a counterparty failure

Start with deposits as a state variable rather than a label. Its role is customer funding and payment claims. Apply a counterparty failure. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include direct exposure, collateral and network propagation. The native weakness to watch is runoff speed.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is liquidity buffer. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 7: deposits under a cyber incident

Start with deposits as a state variable rather than a label. Its role is customer funding and payment claims. Apply a cyber incident. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include availability, data integrity and recovery sequencing. The native weakness to watch is runoff speed.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is liquidity buffer. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 8: deposits under a data-model drift event

Start with deposits as a state variable rather than a label. Its role is customer funding and payment claims. Apply a data-model drift event. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include forecast error, overrides and recalibration. The native weakness to watch is runoff speed.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is liquidity buffer. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 9: deposits under a collateral haircut increase

Start with deposits as a state variable rather than a label. Its role is customer funding and payment claims. Apply a collateral haircut increase. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include lendable value, margin demand and fire-sale pressure. The native weakness to watch is runoff speed.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is liquidity buffer. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 10: deposits under a liquidity freeze

Start with deposits as a state variable rather than a label. Its role is customer funding and payment claims. Apply a liquidity freeze. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include funding access, payment queues and survival horizon. The native weakness to watch is runoff speed.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is liquidity buffer. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 11: loans under a sudden rise in outflows

Start with loans as a state variable rather than a label. Its role is credit claims on borrowers. Apply a sudden rise in outflows. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include speed, concentration and replacement capacity. The native weakness to watch is default and prepayment.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is capital and pricing. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 12: loans under a market-price shock

Start with loans as a state variable rather than a label. Its role is credit claims on borrowers. Apply a market-price shock. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include mark-to-market effects, haircuts and forced actions. The native weakness to watch is default and prepayment.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is capital and pricing. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 13: loans under an operational outage

Start with loans as a state variable rather than a label. Its role is credit claims on borrowers. Apply an operational outage. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include backlog, manual fallback and reconciliation integrity. The native weakness to watch is default and prepayment.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is capital and pricing. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 14: loans under a policy-rate shift

Start with loans as a state variable rather than a label. Its role is credit claims on borrowers. Apply a policy-rate shift. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include repricing gaps, behavioural response and lag. The native weakness to watch is default and prepayment.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is capital and pricing. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 15: loans under a confidence shock

Start with loans as a state variable rather than a label. Its role is credit claims on borrowers. Apply a confidence shock. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include withdrawal behaviour, spreads and market access. The native weakness to watch is default and prepayment.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is capital and pricing. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 16: loans under a counterparty failure

Start with loans as a state variable rather than a label. Its role is credit claims on borrowers. Apply a counterparty failure. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include direct exposure, collateral and network propagation. The native weakness to watch is default and prepayment.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is capital and pricing. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 17: loans under a cyber incident

Start with loans as a state variable rather than a label. Its role is credit claims on borrowers. Apply a cyber incident. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include availability, data integrity and recovery sequencing. The native weakness to watch is default and prepayment.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is capital and pricing. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 18: loans under a data-model drift event

Start with loans as a state variable rather than a label. Its role is credit claims on borrowers. Apply a data-model drift event. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include forecast error, overrides and recalibration. The native weakness to watch is default and prepayment.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is capital and pricing. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 19: loans under a collateral haircut increase

Start with loans as a state variable rather than a label. Its role is credit claims on borrowers. Apply a collateral haircut increase. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include lendable value, margin demand and fire-sale pressure. The native weakness to watch is default and prepayment.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is capital and pricing. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 20: loans under a liquidity freeze

Start with loans as a state variable rather than a label. Its role is credit claims on borrowers. Apply a liquidity freeze. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include funding access, payment queues and survival horizon. The native weakness to watch is default and prepayment.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is capital and pricing. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 21: reserves under a sudden rise in outflows

Start with reserves as a state variable rather than a label. Its role is central-bank settlement balances. Apply a sudden rise in outflows. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include speed, concentration and replacement capacity. The native weakness to watch is intraday depletion.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is payment throughput. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 22: reserves under a market-price shock

Start with reserves as a state variable rather than a label. Its role is central-bank settlement balances. Apply a market-price shock. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include mark-to-market effects, haircuts and forced actions. The native weakness to watch is intraday depletion.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is payment throughput. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 23: reserves under an operational outage

Start with reserves as a state variable rather than a label. Its role is central-bank settlement balances. Apply an operational outage. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include backlog, manual fallback and reconciliation integrity. The native weakness to watch is intraday depletion.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is payment throughput. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 24: reserves under a policy-rate shift

Start with reserves as a state variable rather than a label. Its role is central-bank settlement balances. Apply a policy-rate shift. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include repricing gaps, behavioural response and lag. The native weakness to watch is intraday depletion.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is payment throughput. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 25: reserves under a confidence shock

Start with reserves as a state variable rather than a label. Its role is central-bank settlement balances. Apply a confidence shock. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include withdrawal behaviour, spreads and market access. The native weakness to watch is intraday depletion.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is payment throughput. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 26: reserves under a counterparty failure

Start with reserves as a state variable rather than a label. Its role is central-bank settlement balances. Apply a counterparty failure. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include direct exposure, collateral and network propagation. The native weakness to watch is intraday depletion.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is payment throughput. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 27: reserves under a cyber incident

Start with reserves as a state variable rather than a label. Its role is central-bank settlement balances. Apply a cyber incident. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include availability, data integrity and recovery sequencing. The native weakness to watch is intraday depletion.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is payment throughput. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 28: reserves under a data-model drift event

Start with reserves as a state variable rather than a label. Its role is central-bank settlement balances. Apply a data-model drift event. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include forecast error, overrides and recalibration. The native weakness to watch is intraday depletion.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is payment throughput. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 29: reserves under a collateral haircut increase

Start with reserves as a state variable rather than a label. Its role is central-bank settlement balances. Apply a collateral haircut increase. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include lendable value, margin demand and fire-sale pressure. The native weakness to watch is intraday depletion.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is payment throughput. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 30: reserves under a liquidity freeze

Start with reserves as a state variable rather than a label. Its role is central-bank settlement balances. Apply a liquidity freeze. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include funding access, payment queues and survival horizon. The native weakness to watch is intraday depletion.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is payment throughput. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 31: capital under a sudden rise in outflows

Start with capital as a state variable rather than a label. Its role is loss-absorbing resources. Apply a sudden rise in outflows. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include speed, concentration and replacement capacity. The native weakness to watch is loss accumulation.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is balance-sheet capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 32: capital under a market-price shock

Start with capital as a state variable rather than a label. Its role is loss-absorbing resources. Apply a market-price shock. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include mark-to-market effects, haircuts and forced actions. The native weakness to watch is loss accumulation.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is balance-sheet capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 33: capital under an operational outage

Start with capital as a state variable rather than a label. Its role is loss-absorbing resources. Apply an operational outage. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include backlog, manual fallback and reconciliation integrity. The native weakness to watch is loss accumulation.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is balance-sheet capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 34: capital under a policy-rate shift

Start with capital as a state variable rather than a label. Its role is loss-absorbing resources. Apply a policy-rate shift. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include repricing gaps, behavioural response and lag. The native weakness to watch is loss accumulation.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is balance-sheet capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 35: capital under a confidence shock

Start with capital as a state variable rather than a label. Its role is loss-absorbing resources. Apply a confidence shock. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include withdrawal behaviour, spreads and market access. The native weakness to watch is loss accumulation.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is balance-sheet capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 36: capital under a counterparty failure

Start with capital as a state variable rather than a label. Its role is loss-absorbing resources. Apply a counterparty failure. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include direct exposure, collateral and network propagation. The native weakness to watch is loss accumulation.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is balance-sheet capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 37: capital under a cyber incident

Start with capital as a state variable rather than a label. Its role is loss-absorbing resources. Apply a cyber incident. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include availability, data integrity and recovery sequencing. The native weakness to watch is loss accumulation.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is balance-sheet capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 38: capital under a data-model drift event

Start with capital as a state variable rather than a label. Its role is loss-absorbing resources. Apply a data-model drift event. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include forecast error, overrides and recalibration. The native weakness to watch is loss accumulation.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is balance-sheet capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 39: capital under a collateral haircut increase

Start with capital as a state variable rather than a label. Its role is loss-absorbing resources. Apply a collateral haircut increase. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include lendable value, margin demand and fire-sale pressure. The native weakness to watch is loss accumulation.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is balance-sheet capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 40: capital under a liquidity freeze

Start with capital as a state variable rather than a label. Its role is loss-absorbing resources. Apply a liquidity freeze. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include funding access, payment queues and survival horizon. The native weakness to watch is loss accumulation.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is balance-sheet capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 41: collateral under a sudden rise in outflows

Start with collateral as a state variable rather than a label. Its role is pledged assets. Apply a sudden rise in outflows. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include speed, concentration and replacement capacity. The native weakness to watch is valuation and haircuts.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is secured funding capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 42: collateral under a market-price shock

Start with collateral as a state variable rather than a label. Its role is pledged assets. Apply a market-price shock. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include mark-to-market effects, haircuts and forced actions. The native weakness to watch is valuation and haircuts.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is secured funding capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 43: collateral under an operational outage

Start with collateral as a state variable rather than a label. Its role is pledged assets. Apply an operational outage. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include backlog, manual fallback and reconciliation integrity. The native weakness to watch is valuation and haircuts.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is secured funding capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 44: collateral under a policy-rate shift

Start with collateral as a state variable rather than a label. Its role is pledged assets. Apply a policy-rate shift. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include repricing gaps, behavioural response and lag. The native weakness to watch is valuation and haircuts.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is secured funding capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.

Laboratory 45: collateral under a confidence shock

Start with collateral as a state variable rather than a label. Its role is pledged assets. Apply a confidence shock. The first task is to write the before-and-after state and identify whether the disturbance changes a stock, a flow, a price, a probability, a timing assumption or several at once. For this combination, the most informative observables include withdrawal behaviour, spreads and market access. The native weakness to watch is valuation and haircuts.

Now close the loop. Ask what management, customers, counterparties, infrastructure or authorities do in response, how long that response takes, and which new state variable carries the consequence forward. The return variable here is secured funding capacity. A strong model must state a failure threshold, a recovery path and an observation that would prove the assumed relationship wrong. If the model records the shock but not the adaptive response, it is open-loop description rather than closed-loop analysis.