Almost nothing you believe about money survives contact with the plumbing.
The money in your account isn’t sitting anywhere. Your bank doesn’t have a box with your name on it. When you pay someone at another bank, nothing travels between them. And the moment your card is approved isn’t the moment anybody gets paid; those are days apart, and the gap is where most of the industry lives.
None of that is a scandal. It’s a set of design decisions, made over 3 centuries, mostly by people solving a specific problem in front of them. This page is the machine those decisions built.
If you already know what a bank’s balance sheet looks like, start at How a payment settles inside one country.
Money is a record, not a thing
Start with the single idea everything else rests on.
When you hold a banknote, you hold a claim on a central bank. When you hold a bank balance, you hold a claim on that bank: an IOU, legally a debt the bank owes you. Neither is a thing you possess so much as a promise you’re owed.
This isn’t a modern abstraction. It’s the oldest arrangement in the subject.
Goldsmiths, and the receipt that became the money
Seventeenth-century London goldsmiths had strongrooms, so people deposited coin with them and took a receipt. The receipts were easier to carry than coin, so people began settling debts by handing over the receipt instead of collecting the coin.
The moment that happened, the receipt was the money. And the goldsmith noticed something: on any given day, only a fraction of depositors came to collect. The rest of the coin sat there. So he lent it out, and wrote more receipts than he had coin.
Wider than the screen; scroll it sideways.
Every bank since works this way. The receipts circulate, the coin doesn’t, and the system depends on not everyone asking at once.
The two-tier map
Here’s the picture the whole course hangs on.
The public tier: you, businesses, most institutions. You hold claims on commercial banks. Your balance is your bank’s IOU to you.
The bank tier: the banks themselves hold accounts at the central bank. The money in those accounts is called reserves, and it’s the only money banks use to settle with each other.
Wider than the screen; scroll it sideways.
Hold that picture. Every mechanism below(cards, wires, netting, stablecoins) is a different answer to the same question: how do the two tiers stay in step?
Your bank shows £500. In what sense do you “have” £500?
You have a claim on your bank for £500, a debt it owes you, recorded as a liability on its balance sheet. There’s no box of notes with your name on it. The bank holds a much smaller quantity of reserves and cash against all its deposits together, which is fine precisely because not everybody asks at once.
What a bank actually is
A bank isn’t a vault. It’s a balance sheet, and reading one explains most of banking.
Double entry: every move has two legs
This is the oldest reliability mechanism in finance and the reason the whole system can be checked. Money never appears or vanishes; it only moves between accounts. Every mechanism further down posts to a ledger like this and has to leave it balancing.
Wider than the screen; scroll it sideways.
Assets are what the bank owns: loans it has made, bonds, reserves at the central bank. Liabilities are what it owes: your deposits, mostly. Equity is the difference: the owners’ stake, and the cushion that absorbs losses before depositors feel them.
Loans create deposits
This one reliably surprises people. Banks don’t lend out deposits they received. When a bank makes a loan, it writes two entries: a new asset(your loan) and a new liability(a new deposit in your account).
Settlement: when a payment becomes real
Everything before settlement(the authorisation, the message, the pending line in your app) is a promise that money will move. The distinction between “promised” and “settled” is where nearly all payment risk lives, and it’s the distinction the rest of this page keeps returning to.
Wider than the screen; scroll it sideways.
Repo: a loan dressed as a sale
Banks lend to each other constantly, usually overnight, and mostly not by lending at all.
Wider than the screen; scroll it sideways.
The difference in the two prices is the interest. The securities are the collateral, and dressing the loan as a sale is what makes it easy to enforce: if the borrower fails, the lender already owns the bonds outright rather than having to claim them.
This market is enormous and nearly invisible, and it’s where a surprising number of crises have started, because when lenders suddenly want more collateral for the same cash, every borrower has to sell something on the same morning.
What a central bank actually controls
A central bank doesn’t set the interest rate on your mortgage. It sets the rate at which banks trade reserves with each other overnight, and everything else is priced off that.
Wider than the screen; scroll it sideways.
A bank makes a £200,000 mortgage loan. Where did the money come from?
Nowhere. It was created. The bank writes a £200,000 asset(the loan) and a £200,000 liability(the deposit) simultaneously. The constraint on doing this is capital requirements and finding borrowers who will repay, not a stock of existing deposits waiting to be lent.
How a payment settles inside one country
Now the machines. A country’s payment systems all do the same job, moving value between banks, and each one makes a different trade between speed, cost and size.
Clearing isn’t settlement
Nearly every confusion about payments dissolves once these two are held apart.
RTGS: the heavy rail
The most important machine is RTGS, for Real-Time Gross Settlement. Unpack the name: real-time means each payment settles the moment it’s submitted; gross means one by one, not bundled; settlement means final.
RTGS moves reserves at the central bank, the top tier of the pyramid in action. It carries enormous values at low volumes: interbank transfers, securities settlement, large corporate payments. Not your coffee.
Settling everything instantly and individually is safe, and it’s expensive: each bank has to hold enough reserves to cover each payment as it goes out.
Batch rails: paying less for liquidity
The other approach is deferred net settlement: collect payments all day, work out the net position of each bank against every other, and settle only the differences at the end.
As you can see, the saving is dramatic. Banks can exchange enormous gross values and settle a small fraction in reserves. But between the promise and the settlement, everyone is exposed to everyone. If a bank fails at 4 o’clock, the payments it “made” that morning were never real.
Cards, which are a different machine entirely
A card tap isn’t a payment. It’s a request for authorisation travelling through four parties.
Wider than the screen; scroll it sideways.
The cardholder’s bank(the issuer) approves against your available credit, the merchant’s bank(the acquirer) gets the merchant paid, and the network in the middle routes messages and sets the rules. The money settles a day or more later through a batch rail underneath.
One language for every rail
Every rail above needs to say the same things: who is paying, who is being paid, how much, why. For decades each system said them differently, in terse formats designed when bandwidth was expensive.
Wider than the screen; scroll it sideways.
The industry is migrating to a single structured standard across domestic and cross-border rails alike. The interesting part isn’t the format. It’s that structured fields make things checkable. A free-text field holding a name and address can’t be validated; separate fields for each can be. Most of the compliance friction in cross-border payments comes from information that arrived as prose.
Instant payment rails settle in seconds. Why hasn’t that replaced the batch rails?
Because instant settlement means each bank has to hold enough liquidity to cover every payment the moment it goes out, around the clock including weekends. Batch netting lets banks settle a fraction of gross value once a day. Instant rails buy certainty and pay for it in liquidity; batch rails buy cheap liquidity and pay for it in risk between clearing and settlement.
How money crosses a border
Here’s the thing nobody tells you: there is no international payment system.
There’s no global ledger and no central bank of the world. So a cross-border payment isn’t a transfer at all. It’s a relay.
The correspondent relay
Banks open accounts with each other. A bank in one country holds an account at a bank in another, and payments hop along the chain of these relationships.
Wider than the screen; scroll it sideways.
Every hop is a real bank doing real work, taking a fee, applying its own compliance checks, and adding delay. That’s why sending money abroad costs what it costs: not one greedy institution, but a chain of them.
The message isn’t the money
Banks coordinate through messaging networks. A message says pay this person this amount. It doesn’t carry value; the value moves separately, through accounts the two banks hold.
Herstatt, and both-or-neither
In 1974 a German bank called Herstatt was closed by regulators at the end of the German business day. Counterparties had already paid it deutschmarks that morning. They were expecting dollars back when New York opened. The dollars never came.
Wider than the screen; scroll it sideways.
The lesson was permanent: if the two legs of a trade settle at different times, somebody is exposed in between. The fix is a family of mechanisms with a shared shape. Both legs settle or neither does.
Why does a £50 remittance sometimes cost £8 and take 3 days?
Because it’s a relay, not a transfer. Each correspondent in the chain charges a fee, applies its own compliance checks, and settles on its own schedule. The FX conversion carries a spread away from the mid-market rate. The cost and delay are the chain, not one institution.
Markets: raising money, and owning things
Above the payment rails sit the markets. Everything here reduces to two ways of raising money, and one long argument about who actually owns what.
Two doors: equity and debt
That ordering, debt before equity, is the backbone of corporate finance, and it comes back as the loss-absorbing stack in securitisation and again in clearing house default waterfalls.
A bond is a loan you can trade
A bond promises fixed payments on fixed dates. Its price and its yield move in opposite directions, mechanically: the payments are fixed, so paying more for them means earning less.
You don’t own your shares
This one is genuinely strange. When you buy a share through a broker, your name is almost certainly nowhere near the company’s register.
Wider than the screen; scroll it sideways.
The register names a central securities depository, which recognises a custodian, which recognises your broker, which has a record saying you have a claim on some of what it holds. You own an entry at the end of a chain of intermediaries.
How a price forms
A price isn’t announced. It emerges from an order book: buyers stacked by what they will pay, sellers by what they will accept, matched by rules.
Wider than the screen; scroll it sideways.
Market makers quote both sides continuously and earn the spread between them, carrying inventory risk for the privilege of being the one always willing to trade.
Where securities are born
Everything above is the secondary market, people trading things that already exist. Securities enter the world through a primary issuance, and the mechanism is usually an auction.
Governments fund themselves this way, week after week, and the auction produces a number the whole market watches: the ratio of what was bid to what was offered. A thin auction is a signal about demand for a government’s debt long before it’s a headline about interest rates.
Why does a bond’s price fall when interest rates rise?
The bond’s payments are fixed and contractual. If newly issued bonds pay more, the only way an existing bond can compete is to become cheaper to buy: the same fixed payments bought at a lower price produce a higher yield. Price and yield are two views of one number.
Leverage, collateral, and the machinery that catches falls
Leverage is the amplifier. Clearing is the safety net. They’re the same story told from two ends.
Margin: the amplifier
At 10 times leverage, a 10% adverse move erases the entire stake. Leverage doesn’t change expected returns; it scales both directions, and it introduces a way to be right eventually and bankrupt first.
Collateral in motion
Positions are marked to market continuously. When one moves against you, you post variation margin: more collateral, today. That call is what turns a paper loss into a cash demand, and it’s why forced selling clusters: everybody gets the call on the same bad day.
The clearing house
A central counterparty steps into the middle of every trade: it becomes buyer to every seller and seller to every buyer. Instead of everybody carrying exposure to everybody, everybody carries exposure to one heavily-regulated institution.
That institution needs a plan for a member failing, agreed in advance.
Manufacturing safety out of risk
Take 1,000 mortgages, pool them, and sell claims on the pool. Now slice the claims by who absorbs losses first.
The engineering is real: from a pool of individually risky loans you can genuinely manufacture a senior claim that’s safe under most conditions. The 2008 failure wasn’t that the machine didn’t work. It was that the inputs were worse than represented and the loans turned out to fail together rather than independently, and a top tranche that assumed independence has no defence against everything going wrong at once.
Why does a clearing house reduce risk, given the total exposure hasn’t changed?
It replaces a dense web of bilateral exposures with a hub. Positions net down enormously, one institution’s risk management applies to everyone, and the order of loss absorption is agreed in advance rather than argued about during a crisis. The exposure is concentrated, not removed, which is why clearing houses themselves are so heavily regulated.
Money on a shared ledger
Now the rebuild. Everything above evolved over centuries in layers. The question this part asks is what happens when you build it again, on a ledger many parties can write to.
Collapsing two tiers into one
Wider than the screen; scroll it sideways.
The mechanics are older than they look. The issuer takes a dollar, holds it in reserve, and issues a token. Tokens are created only against money received and destroyed when money is returned. That’s mint and burn.
Wider than the screen; scroll it sideways.
The legal cage
A token is code, but the claim behind it is legal. What actually protects a holder is the structure around the issuer: which entity holds the reserves, what they may be invested in, whether they’re segregated from the issuer’s own money, and whether holders can redeem at par.
Wider than the screen; scroll it sideways.
How a peg actually holds, and how it breaks
The peg is held by arbitrage. If the token trades below a dollar, someone buys it cheap and redeems it for a dollar; if above, someone deposits a dollar and mints. The profit closes the gap.
That loop depends entirely on redemption actually working. When a peg breaks, it is almost never the code. It’s doubt about the reserves, or a redemption channel that has stopped functioning.
Not every token is the same kind of thing
“Digital money” covers at least three arrangements that differ in who owes you.
Wider than the screen; scroll it sideways.
A stablecoin is a claim on a private issuer holding reserves. A tokenised deposit is an ordinary bank deposit that happens to move on a shared ledger; your claim is still on your bank, with whatever protection that carries. A central bank digital currency is a claim on the central bank itself: the top tier of the pyramid, in token form.
Retail CBDCs remain mostly unbuilt, and the reason is political rather than technical: the launched ones work. Nobody has made the case for one convincingly enough to carry the cost of a central bank holding accounts for the public.
What actually gets tokenised
Wider than the screen; scroll it sideways.
Beyond dollars, the assets moving onto shared ledgers first are the ones where the existing settlement machinery is slowest and the ownership chain longest: short-dated government paper, money market funds, private credit, gold.
A token trades at $0.97. What has the market decided?
That redemption is doubtful. If anyone could reliably redeem it for a dollar, buying at $0.97 would be free money and the gap would close. A sustained discount is the market pricing the probability that the reserves aren’t there, or can’t be reached.
Token markets, and what actually changed
The last part rebuilds the markets from part five on shared ledgers, then asks honestly what improved.
Trading against a formula
An automated market maker replaces the order book with a pool and an equation. Anyone can supply assets to the pool; the formula quotes a price that moves as the pool’s balance changes.
It’s a genuinely new market structure, always available, with no counterparty to find. It also has a cost that’s invisible until you measure it.
Putting the liquidity where the trading is
A plain pool spreads its assets across every price from zero to infinity, and almost all of it sits at prices that will never trade. Concentrated liquidity lets a provider choose a band instead.
Inside the band you earn far more fees for the same capital. Outside it you earn nothing and hold entirely the wrong asset. It converts a passive position into an active one that needs managing, which is a familiar trade: it’s market making from part five with the inventory risk made explicit.
Liquidation, mechanised
Margin from part six, in software. Positions are watched continuously against a mark price deliberately built to resist manipulation, and closed automatically when collateral runs short. Cascades happen when liquidations push prices further in the direction that triggered them.
Not every trade happens in public
Wider than the screen; scroll it sideways.
A large order shown to a public book moves the price against itself before it fills. So large trades have always had somewhere quieter to go: venues that don’t display orders, or a request for quote flow where you ask a handful of dealers privately and pick the best answer.
Wider than the screen; scroll it sideways.
Both patterns have been rebuilt on shared ledgers, which is a useful check on the claim that everything is now transparent. The demand for privacy in large trades is structural, not a legacy of old market plumbing, and it reappears the moment size does.
Both-or-neither, again
Two ledgers that can’t see each other still need to trade without either side going first. The early answer was bridges(hold assets on one chain, issue claims on another), and bridges have been the single largest source of losses in the sector.
The better answer is a construction where both legs complete or neither does: two conditional locks and one shared secret, no custodian in the middle.
Wider than the screen; scroll it sideways.
The treasury desk becomes software
Wider than the screen; scroll it sideways.
Put the pieces together and something genuinely new appears. A corporate treasury holding several kinds of money(bank deposits, money market funds, tokenised cash, stablecoins) on ledgers that settle in seconds can rebalance continuously rather than once a day at a cutoff.
The honest scorecard
Some things genuinely improved: settlement that used to take days can complete in seconds, collateral can move as fast as a message, and a treasury operation that needed a desk of people can run as software.
Some things didn’t change at all. Somebody still has to hold the reserves. Somebody is still accountable when it breaks. Credit risk didn’t disappear because the ledger is shared; it moved. And a good deal of what is described as decentralised has an issuer, an administrator, or a multisig that can freeze your balance.
Atomic settlement removes the need for a clearing house. True?
No. Atomic settlement removes settlement risk, the risk that one leg completes and the other does not. It does nothing about counterparty risk before settlement, or about netting down exposures, or about what happens when a participant fails with open positions. Those are the problems clearing houses exist for, and they survive the technology change.
Where this leaves you
The machine isn’t complicated so much as layered. Records rather than objects. Two tiers that have to stay in step. Clearing separated from settlement. Both-or-neither wherever two legs could come apart. Collateral, and a written order for who absorbs losses. And, most recently, an attempt to rebuild all of it on ledgers many parties can write to, which has re-derived several of the same answers, sometimes the hard way.
If you’ve followed this far, you already read financial news differently.