Batched, One by One, or Instant: Settlement Compared

Three settlement shapes exist, and three questions separate them: whether instructions are gathered and netted before anything settles or each is settled on its own, whether the payee is credited before that settlement or after it, and whether the arrangement runs at every hour. Each hour, interval and value limit is set by the Reserve Bank of India and moves, so the shapes outlast the values.

Most people have used all three of these this month without being able to name one of them. Money sent went out and turned up somewhere later. Money somebody else sent appeared in the account before the message announcing it had been read to the end. A large payment an employer made sat somewhere for a stretch and then landed. Three different arrangements sit underneath those three experiences, and the difference between them is not speed. Speed gets noticed. Speed is a consequence rather than a design.

Here is what actually differs. Gathering instructions and settling only the difference between what each bank owes each other is cheap, and it is slow because nothing moves until the gathering stops. The expensive thing gets done once for every single instruction, so settling each instruction on its own is immediate and expensive. Crediting the payee before the two banks have squared up feels immediate to everybody watching, and it moves the waiting onto a bank instead of removing it. Cost, speed and who does the waiting are the three trades, and every settlement arrangement is a set of answers to them rather than a thing of its own kind.

The claim that every arrangement is only a set of answers is the durable part. Names change with the country and with the decade. Behaviour does not. A reader holding the three questions can have an arrangement described that was never heard of before, in a country never visited, and place it within about four sentences of the description.

What are the three settlement shapes, described by what they actually do?

The order that makes them stick is the amount of waiting each one contains, so take them that way.

GATHERED AND NETTED. Instructions pile up. Nothing settles while they pile. At the end of the gathering, everything in the pile is set against everything else until only a net position for each bank against each other bank is left standing, and one movement of money discharges the whole batchThe set of instructions collected between one settlement and the next. A batch is a grouping, not a length of time, and how big it gets is what the control below moves.. Think of ten shopkeepers in one lane who buy from each other all day and square up once in the evening, with only the differences changing hands. Almost nothing physically moves, and everybody's position is correct.

SETTLED ONE AT A TIME. Each instruction is discharged on its own and in full, as it arrives. Nothing is set against anything. If the same two banks send each other a hundred instructions, that is a hundred separate discharges rather than one. Think of the same ten shopkeepers insisting on cash across the counter for every single item. Nobody is ever owed anything for a moment longer than it takes to hand the notes over, and the counting takes all day.

CREDITED FIRST AND SETTLED AFTERWARDS. The payee's bank puts the money into the payee's account straight away, on the strength of the instruction alone, and squares up with the payer's bank on some other schedule of their own. Think of the shopkeeper who hands over the goods because the customer's brother is standing there, and settles with the brother later. Gathering and netting, settling one at a time, and crediting before settling place any arrangement anybody names, in any country, without its name being known.

What three questions place any arrangement?

The three shapes above are useful and they are not the apparatus. The apparatus is smaller and it fits on one hand. Ask these three, in this order.

ONE: netted in a batch, or each instruction on its own? This is a question about arithmetic. Is the amount that moves the total of everything, or the difference after everything has been set against everything else?

TWO: is the payee credited before the settlement or after it? This is a question about who is exposed. If the credit comes first, some party has parted with money it has not yet been given.

THREE: does the arrangement run at every hour? This is a question about availability, and it looks like the softest of the three. Availability is not the softest of the three, and the reason is coming.

The three questions are independent of one another, so they place far more arrangements than the three most people have met. Three questions with two answers each give eight combinations. The credit-first shape can sit on top of either arithmetic, so the three shapes above account for four of them. The remaining four are not exotic; some of them exist, some of them are being built, and all of them are readable the moment the question stops being what something is called and becomes these three things about it.

What does gathering instructions into a batch buy, and what does it cost?

What gathering buys is the easy part, and nobody argues with it, so start there. A settlement between two banks is an expensive event. A settlement uses somebody's money on the day, it uses somebody's attention, and it has to be right. If one settlement can discharge fifty thousand instructions, the expensive thing has been done once instead of fifty thousand times. Doing the expensive thing once instead of fifty thousand times is the entire argument for gathering, and it is a good argument.

Now the cost. Most explanations stop early here and leave a reader believing that netting is free. From the point at which an instruction is taken in to the point at which its batch is discharged, that instruction is an undischarged obligationAn amount one party owes another that has been agreed and recorded but not yet paid over. The obligation is real, it is enforceable, and it sits on somebody's books until the payment happens.: a real amount that one bank owes another, agreed and not yet paid. And it is not sitting there alone. Every other instruction accepted since the last settlement is sitting there with it.

The bigger the batch, the fewer the settlements and the more that has piled up waiting for each one, and those two are the same lever pulled from opposite ends. Notice the precise shape of the second half, because a loose version of it is easy to write and wrong. The pile is not the same size all the time. Immediately after a settlement it is empty. The pile fills as instructions arrive, and it reaches the batch size at the moment the next settlement lands. So the batch size is the peak of the pile rather than its permanent level, and the peak is what any arrangement has to be built to survive.

The lane of shopkeepers again. If they square up once a week instead of once an evening, they save six squaring-ups and each of them is carrying six days of everybody else's promises by Saturday night. Nothing was created and nothing was destroyed. The waiting was concentrated.

What happens when a payee is credited before the banks have settled?

Crediting first and settling afterwards is the shape most people use daily and the fewest could describe, so it is worth going slowly. The payee's bank receives an instruction saying that money is coming. The payee's bank does not wait to be paid. The bank puts the amount into the payee's account there and then, and squares up with the payer's bank afterwards on whatever schedule the two of them run on.

The payee has the money and the payee's bank is out of pocket until that settlement happens, so the wait was not removed, it was moved onto a bank where nobody outside can see it. That is the trick, and calling it a trick is not an accusation. Moving the wait onto a bank is a genuine service. The bank lends its own position for the length of the gap, in exchange for being an arrangement people want to use. The amount it has to keep free to do that, all through the working stretch, is its intraday liquidityMoney a bank keeps available during the working stretch to meet payments it has to make before the money coming the other way has arrived. Intraday liquidity is covered separately and only named here., and that is a real cost carried by a real party.

One sentence carries the whole of it. When the screen says done, in this shape, the money really has arrived and the settlement still has not happened. Both of those are true at once, and neither cancels the other. The payee is not being deceived and the payment is not finished. There is simply a gap, and the payee is standing on the far side of it looking at a completed transfer.

Why does running at every hour force a particular answer to the other questions?

Question three looked like the soft one. Question three is the one with teeth, and here is why. A settlement between two banks is not a thing a machine does alone in the dark. A settlement needs the parties that stand behind it to be functioning. The whole promise such an arrangement made was that it would not make anybody wait, so an arrangement that answers at three in the morning cannot wait for a working stretch to credit anybody.

So it has to credit first. There is no third option. The credit happens on the strength of the instruction, and the squaring up happens later, and somebody has agreed to be out of pocket in between. Running at every hour is not a feature bolted onto a settlement arrangement, it is a decision that somebody will carry an obligation overnight, and the arrangements that run at every hour are the ones that found a party willing to.

Read that backwards and a great deal that otherwise looks arbitrary comes clear. The overnight obligation explains why an arrangement of this kind has limits on it at all. The obligation explains why the arrangement can stop when something upstream is wrong. The obligation explains why the party carrying the gap cares a great deal about how long the gap is, and about how big the pile inside it gets. None of those are malfunctions. The limits and the stoppages are the visible edges of an obligation somebody agreed to carry so that the payee would not have to wait.

What has happened when a payer is debited and the payee is not credited?

Two questions are worth asking when a payer has been debited and the payee has not been credited, and everything else people say about the situation is noise.

What state is the instruction in? The payer's bank has acted on it. The payee's bank has not completed its side. So the amount is at a point between those two events rather than nowhere. The amount is in transitThe state an amount is in when one party has acted on an instruction and the other has not completed its side. In transit is a recorded state on somebody's books, not a gap in the record., a recorded state with a name, and it is somewhere. Every arrangement runs a reconciliationThe routine matching of what one party recorded against what the other recorded, which finds anything sitting in an incomplete state and resolves it. Reconciliation runs on its own schedule whether or not anybody has complained. that finds instructions sitting in that state and resolves them, and it runs on its own whether or not anybody has said a word.

Who does the payer take this to? Their own bank, and nobody else in the chain. The payer's own bank acted on the instruction, holds its own record of having acted, and is therefore the only party with a view of where the amount currently stands.

The amount is not lost while it is in that state, and knowing the state exists is what makes the route worth taking rather than worth dreading. The route itself, and every period allowed at every step of it, is set by the Reserve Bank of India. Nothing a payer does produces this state, and nothing a payer could have done differently prevents it. In transit is a state the arrangement can be in.

Why does the average payment decide which shape a system ends up with?

Setu Payments Limited, an invented payment system, reports two figures for the stated year and nothing else: 1,200 crore instructions, carrying Rs 3,60,000 crore of value. Put the value over the count, work it through, and the average instruction comes to Rs 300.00/-. Everything below falls straight out of that division, so it is worth doing rather than reading.

A division result is the easiest thing to over-read, so one caution before that figure does any work. Rs 300.00/- is what falls out when one reported total is divided by another reported total. The average is a quotientThe result of one division. A quotient describes the relationship between the two numbers that produced it and does not describe any individual item inside either of them., so it describes how those two totals stand to each other and describes nobody's actual payment. Two systems that are nothing alike land on the identical figure: one where nearly every instruction really is about that size, and another where a handful of very large ones sit inside a vast crowd of tiny ones. The record carries one count and one value and nothing behind either, so there is no middle instruction, no spread and no picture of how the instructions are distributed.

Now the consequence. Settling each of those 1,200 crore instructions on its own means doing the expensive thing 1,200 crore separate times, against Rs 300.00/- of value each time. The settlement cost lands on whatever value passed through it, and Rs 300.00/- is not much value to land it on. An average of Rs 300.00/- pulls towards gathering and netting for exactly that reason. The only way one settlement can earn its keep is to spread it over an enormous count of instructions rather than a handful.

Take the other end and hold the value still. The same Rs 3,60,000 crore, moving at an average of Rs 3,00,000/- a payment, needs 1.2 crore instructions. The instruction count is a thousand times smaller and the average is a thousand times larger, and the reason is the same in both directions: the Rs 3,60,000 crore carried never moved. Now the expensive thing is trivial beside the amount it is discharging, and the pile waiting inside a batch would be made of payments that size rather than of small ones. The average payment does not merely suit a shape, it selects one, and it selects opposite shapes at the two ends.

What moves when more instructions are gathered into each batch?

Description gets a reader only so far. The control below carries one specific thing that is hard to believe until it has been moved. Gathering and netting is cheap. Settling one at a time is immediate. Filing cheap gathering and immediate settling as two arrangements is natural and wrong. The two are the ends of one lever, and there is nothing between them except a choice about where to stand.

One answer worth settling before the control moves.

Two things to notice, and the second one is the payoff. First, the default. At 12 crore instructions in a batch, the year's 1,200 crore instructions need 100 settlements, and 12 crore instructions are sitting undischarged when each of those settlements arrives. One division gives it: 1,200 crore divided by 12 crore is 100. The number waiting at the moment of settlement is the batch size itself, so the second reading needs no arithmetic at all.

Second, the two ends. At the bottom of the control there is one settlement for each instruction: 1,200 crore settlements in the year, with a single instruction waiting at any moment. At the top there is one settlement for the whole year, with 1,200 crore instructions waiting for it. The two ends are not two kinds of arrangement but gathering and netting on one side and settling one at a time on the other, reached by moving one control. Everything anybody actually runs sits somewhere between them, and the closest the two readings ever come on this ladder is at a batch of 1,20,000 instructions, where 1,20,000 are waiting against 1,00,000 settlements in the year. Nowhere does any setting give few settlements and little waiting at the same time.

How does anybody actually use these three questions?

Who sets every hour, interval and limit that separates these in practice?

A reader comes to a comparison wanting four numbers. How long each arrangement stays open. How often a batch is settled. How much a single instruction may carry. What may be charged for one. Every one of those is set by the Reserve Bank of India, and every one of them is revised.

A remembered interval would not be merely stale on the day it changed, it would be wrong, and it would send somebody to a counter with a false expectation on the one day it mattered to them. So the four rows carry their labels, the authority is printed inside each one, and the value is left out. The sheet is meant to be carried to the address printed in it, with the middle column filled in there. Remembering that a ceiling exists and who moves it lasts a great deal longer than remembering the number that sat under it last year, so a blank column still works.

A comparison is exactly where a wrong number does the most damage: it does not just mislead, it ranks.

One last idea before the close, and it is the one that makes all of this durable. Ask when a settlement reaches finalityThe point after which a completed settlement is not going to be undone, whatever happens to the parties afterwards. When that point arrives in any given arrangement is settled by rule rather than by anybody's preference.: the point past which it is not going to be undone. Different arrangements answer the finality question differently, and finality is the deep reason the three questions matter. Everything about this subject is a set of choices about how much time passes between a promise and a point of no return, and about who is standing in that gap. When an arrangement is next described in terms of how many seconds it takes, that is the question to put instead.