How to Map Derivative Operational Risks Across a Period

Nothing on this map needs a rate to move. Five tasks make up a period, and each one has a way of not happening: a benchmark reading nobody wrote down, a fraction applied from the wrong field, two systems holding terms that differ, money released on a day it cannot land, an entry that was never made. Each carries its own check and the moment that check has to run.

An agreement, read as a document, is a set of obligations. Read as work, it is something else entirely: a list of tasks, each performed by a named person, on a stated day, out of a stated source. Every failure on this map is one of those tasks not being done, being done late, or being done from the wrong document. None of them waits on a market. Work that waits on no market earns a map of its own.

One thing has to be settled before the stations open. The agreement is given no worth anywhere below. Establishing that worth partway through would take benchmark readings for periods that have not happened yet, and the only reading on hand is the opening one. So where a valuation would ordinarily sit, an empty cell appears with its reason printed inside it.

Two invented sides carry this map. Chitrakoot Cements Limited holds the row nobody rewrites, and 7.20 per cent leaves it annually. Opposite that sits Saranga Capital Limited, whose row is torn up and reissued each period out of whatever the floating benchmark showed on the day it was meant to be read. Both rows are sized against a notional of Rs 1,000 crore. The Rs 1,000 crore notional was never scheduled to travel anywhere, so no operational failure below can put it in play.

Why do these risks get a map of their own?

Almost every other subject in this family turns on a benchmark reading moving. A reading rises and one side's row gets dearer. A reading falls and the difference points the other way. Rate risk is about a market, so its pictures are drawn against a range of readings. The reading is the axis the danger lives on.

Operational failure has no such axis. The trigger is a calendar rather than a market. A person did not walk to the terminal on Tuesday. A field in a spreadsheet was copied from the wrong column. Two systems were loaded from two different documents in two different weeks. A message asking for a change of account arrived on a Friday afternoon and nobody rang back. Rates could have stood dead still through every one of those and each would still have cost money. So the map is laid on the period, station by station, in the order the work is done.

Before the stations, get the scale right. People routinely get it wrong in the frightening direction. The headline on this agreement reads Rs 1,000 crore. The headline is a measuring stick: it sizes both rows and it was never going anywhere. The fixed row gets worked first: 7.20 per cent applied once across the period, against Rs 1,000 crore of notional, produces Rs 72.00 crore. Run the benchmark's opening reading of 6.00 per cent through the same stick and the floating row lands on Rs 60.00 crore. Subtract, and Rs 12.00 crore is the only sum that goes anywhere at all.

So the money an operational failure can reach on this agreement is Rs 12.00 crore, not Rs 1,000 crore. Rs 12.00 crore is not a small relief and not a large one either. The figure is simply the right one, and a map that puts the wrong number against a station sends the checking effort to the wrong place. Scale it down to something a hand can hold: Rs 1.20 out of every Rs 100/- of notional moves across the period. The digits 1.20 repeat the 1.20 percentage points separating the two rates, and the repetition is forced rather than lucky. A percentage already is a per-hundred figure, and has always meant exactly that. The period is counted whole, so nothing shrinks it.

Station one: what happens when the reading is never written down?

The floating row does not carry a number. The row carries an instruction, and the instruction has to be executed: somebody observes the benchmark on the stated day and records what it showed. On this agreement that observation gave 6.00 per cent for the opening period, and every figure downstream of it is built on that one act.

When nobody performs it, the period simply cannot be worked out, and the hole stays invisible until somebody tries. The attempt may come weeks later. The reset day passed quietly, nothing broke, no message arrived, and the absence only announces itself when the compute station reaches for a field that has nothing in it. By then the day the reading was supposed to be taken on has gone, and what the benchmark showed that morning is not something anybody can conjure back at a desk.

The meter outside a house works the same way. Somebody is meant to read it on the fifth. One month nobody walks out, so the bill is estimated; the following month the real reading turns up and the correction is brutal; and by then the standing instructionAn account and route left on file so a repeating payment need not be re-addressed each time. Convenient, and precisely what somebody would want to change. has already pushed two wrong amounts out of the account. The price of a unit of electricity did nothing at all in either month.

There is a second and much quieter version of this failure: the reading is taken from exactly the right source, on the wrong day. A reading taken on the wrong day produces a number. The field fills, the compute station runs, both rows come out, an amount is agreed and money moves. Nothing anywhere signals a problem. A wrong number that looks right travels further than no number at all. The figure goes through the confirm station, through the payment, into the record, and it stays there.

So the check at this station is not a check on the amount. The check is on the act: the reading is set against the instruction written into the agreement on the reset day itself, not later when the payment is being prepared. Timing is the whole of its value. A comparison run at the payment stage can show that a number exists; it cannot show that the number was observed on the day the document names, and by then that morning is unrecoverable.

Station two: how does half the money turn on one field?

The compute station takes the notional, the fixed rate, the reading the previous station produced, and the fraction the period is counted at, and turns those four into two row totals and one difference. The compute station is arithmetic. Nobody thinks of arithmetic as risky, and that is exactly the problem: it is the quickest error on the map to make and the slowest to find.

Watch the size of it. The size is what surprises people. Hold everything steady. The notional stays at Rs 1,000 crore. The fixed rate stays at 7.20 per cent. The opening reading stays at 6.00 per cent. 1.20 percentage points is the whole of the distance between the two rates, and carried across the notional for one full period that distance becomes Rs 12.00 crore. Now change nothing except the fraction. Counted at a fraction of 1.0000, the difference is Rs 12.00 crore. Counted at 0.5000, a fraction declared here purely so the effect is visible and belonging to no market's convention anywhere, the same four inputs give Rs 6.00 crore.

Half the money turned on a field nobody looks at twice. No rate moved. No reading was wrong. The notional was right on both sides throughout. And if one side applied the wrong fraction and the other did not, the two of them are now Rs 6.00 crore apart on a period in which the market did precisely nothing.

How the fraction is arrived at, and why two rows of the same agreement can legitimately count the same stretch of days differently, is covered under the day count convention. The lesson at this station is narrower and more useful: the fraction is an input like any other, it lives in a field, fields get populated from somewhere, and somewhere can be wrong.

The check is a recomputation, and it has a moment. Both rows are worked again, independently, from the agreement rather than from the system that produced them the first time. The moment is before the amount is put to the other side, not after. A figure that has been exchanged picks up an authority it never earned: it is now on a screen at two firms, somebody has acknowledged it, and unwinding it costs a conversation that the person who spotted it will often decide is not worth having.

Station three: what is a confirmation actually comparing?

Each side keeps its own copy of the arrangement in its own systems, loaded by its own people, at its own moment. The confirm station sets those two copies against one another and against the document both sides signed. Setting the copies side by side is the obvious part. The hunt underneath it is not obvious at all.

The confirmation is not hunting for a difference in this period's amount. A difference in the amount is loud. The mismatch shows up as a breakA mismatch between two parties' records of the same arrangement, standing unresolved until somebody accounts for it. the moment the two figures are laid side by side, and somebody is already looking at it. The station exists for the quiet case: two sides agreeing precisely on the figure while holding different terms underneath it.

Two records can agree on the money for this period and disagree on the method the period is counted by, on the rule that decides which day a payment lands on when the calendar is awkward, and on the date the last period ever ends. Every one of those is invisible while the figures match. Agreement on one period conceals all three of them, and each will surface later at a moment of its own choosing.

Which is why the timing of this check is unusual. The check runs when the arrangement is entered, and it runs again every time either side alters a stored term. An amendmentA term rewritten after both sides have already signed. The arrangement carries on; one field inside it now says something different. is the dangerous case precisely because nobody is watching for one: the arrangement was confirmed at the start, the confirmation was filed, and the change came afterwards into a system whose owner had every reason to think the checking was done. Where no golden sourceThe one copy of a term every system is required to agree with. Where no copy has been given that status, two systems disagreeing is nobody's error. has been named, two systems drifting apart is not even anybody's mistake.

And what happens when the two sides simply cannot be brought together is not something to reason out from first principles. The answer is written into the agreement, and finding it is covered under reading the agreement.

Station four: why does the wrong day cost as much as the wrong amount?

One amount moves. Rs 12.00 crore leaves Chitrakoot Cements and reaches Saranga Capital, on a stated day, down a stated route. Two failures live at this station and they look unrelated until a closer look shows they are the same shape.

The first is a payment generated for a date on which the money system is shut. The instruction is perfect, the amount is right, and the day cannot carry a payment at all. Two sides can be looking at two different holiday calendarThe list of days on which a given money system is shut. Two countries keep different lists, so a date that works perfectly for one side can be dead for the other. lists, and a date one of them regards as ordinary can be dead for the other. The agreement carries a rule for exactly this, and the rule is the thing the payment date must be generated from, rather than from a habit or from what last period did.

The second is a payment sent on account details that were changed and never verified. A message arrives, plausibly worded, asking that the net be sent somewhere new. Somebody helpful updates the file. The money leaves down the new route and is gone.

There is a reasonable question about why that belongs on an operational risk map at all rather than under security. A changed account belongs on the map because it has the identical shape to the other four failures: a task performed by a person, from a document, on a day, with nothing about a rate anywhere in it. Splitting it out would leave this map tidier than the work it describes. A map must never be that.

Both checks are about timing and route. The payment date is produced from the agreement's own rule and set against the calendar in advance of the day rather than on it. On the day there is no time left to do anything about the answer. And a change of account details is verified down a route that did not begin with the message asking for the change. Ringing the number printed in the message is not verification; it is the same message, twice. Checking down a second route is the plainest instance of maker checkerAn arrangement of work under which whoever produces a figure is never the person who releases it. The second pair of eyes belongs to somebody with nothing invested in the first pair having been right. earning its keep.

When a payment between two sides has to be made, and the channel it travels through, are set by the Reserve Bank of India at rbi.org.in.

Station five: what does an empty record cost, and when?

Each period leaves behind four things: what was observed, what was computed on each row, what was agreed with the other side, and what actually moved. The fifth station is the act of writing those down. The record station produces nothing anybody wants this week, and that is precisely why it gets skipped.

The cost of skipping it is never paid in the period that skipped it. The delay in the bill is what makes the fifth station hard to argue for. Nothing goes wrong in the period; the money moved, the figures matched, everyone got on with the next thing. The bill arrives much later, when somebody asks a question about any of the four earlier stations and there is nothing to answer it with.

An error found in one period cannot then be traced back to the period that produced it. Worse, a recurring error stops looking like a recurring error. Without a trail, four instances of one cause look like four unconnected incidents, each with its own explanation, each closed on its own. The pattern that would have named the cause on the second occurrence is simply not visible. And an escalationHanding an unresolved item upward to somebody whose job includes deciding it. The word earns its place because an item never handed up is an item that has quietly stopped being anybody's. nobody can support with evidence tends not to happen.

Three things about the keeping of records on a privately agreed arrangement are named on this map and none of them is written out. The three are these: what has to be reported about the arrangement and to whom; what collateral one side places against it; which documents are kept, and the retention periodThe stretch of time a document must be kept before it may be destroyed. Who sets it, and how long it runs, are the authority's to state. attached to each. All three belong to the Reserve Bank of India at rbi.org.in. On the map they appear as named rows with the value column empty, and that emptiness is deliberate.

When does each check actually have to run?

A check named without its moment is half a check. Most of the value in every one of the five is in the timing, and moving a check by a week converts it from something that prevents a loss into something that describes one. Set them out against the clock rather than against the stations.

What does one period look like walked end to end?

Here is the whole routine run once on the invented agreement, with the amount each station produces and the comparison each check makes. Nothing in this walk is new arithmetic; it is the same four inputs carried through five sets of hands.

Notice that only one figure in that whole walk ever leaves anybody's building, and it is Rs 12.00 crore. The Rs 72.00 crore and the Rs 60.00 crore are working figures. The two row totals exist so the difference can be found, they are computed twice by two firms, and neither is ever paid by anybody. A map that put an amount at risk against the fixed row would be pointing at money that was never in motion.

Which risk cannot be priced here at all?

One operational risk on this arrangement arises at no station whatever, and it is the risk of carrying a stale figureA number still in use after the thing it described has moved on. Nothing about it looks wrong; it simply stopped being true and nobody was told. for the agreement's worth today. Nobody performs a task that creates it. The staleness accumulates on its own, in the gap between when a figure was produced and when somebody last asked whether it still held.

No cost can be put on it. Working out the arrangement's worth partway through needs readings for periods that have not arrived, each carrying the date it applies to. The record behind these notes holds one reading, for the opening period, and nothing after it. An invented cost would be indistinguishable from a real one.

So the risk is drawn beside the period rather than on it, and its cost column is left blank with the reason printed inside the cell. The exposure routine applies the same discipline: an empty cell that carries its reason is finished work, not a hole. A cell that carries a plausible number nobody can source is the hole.

Why do the small errors outlive the large ones?

Ranked by how long each one survives, the five failures come out backwards from how they rank by size. A reading nobody took brings the whole period to a halt and gets mended the afternoon it is noticed. The failure is loud, expensive in irritation, and over. The failures that last are the ones small enough that fixing them properly feels disproportionate.

The wedding caterer bills per plate. The hosts count children as half a plate; the caterer counts every seat. Each function ends with a small argument settled by splitting the difference, everybody parts on good terms, and the same argument is waiting at the next function. Nobody wrote down that the two households were counting different things.

Notice what the teaching fraction at station two was doing, and what it was not. A fraction of 0.5000 against 1.0000 was chosen so the mechanism is unmissable: it halves the money and nobody could miss that. Two counting methods in real use do not differ by half; they differ by a sliver. A sliver is precisely what makes a difference settleable. No pair of real methods and no figure for the sliver appears anywhere on this map, so the periods below are drawn with the gap marked and deliberately unpriced. The size of it is not the point. Its persistence is.

The rule the whole map exists to deliver is this: a difference resolved without being explained is a difference that will be back. The number that closes it is not the useful output. The station it came from is, and writing that down is what turns four incidents into one problem somebody can actually fix. Finding the root causeThe thing that produced a difference, as distinct from the difference itself. Closing the second without finding the first leaves the machinery entirely intact. costs an afternoon once; settling the symptom costs an afternoon every period, forever, and nobody ever adds those afternoons up.

What a finished map still cannot say

A finished map says what can fail, where it fails, how much is moving when it does, and what would catch it. Whether either side ought to have entered this agreement is a different question. How hard the checking should be pushed turns on what a side already owes elsewhere, how many of these it is running at the same time, who performs each task, and what else is on those people's desks that week. None of that sits in the record behind these notes.

Chitrakoot Cements Limited and Saranga Capital Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.