How to Maintain a Derivatives Decision Log: Write First

Keep the log for one reason: every entry was set down while the period ahead was still unrun. An entry opens and closes on the decision day. The entry carries the agreed terms, the writer's own reasoning, and each unfillable cell with its reason printed inside. The multiplier the rates work on stays in its own column, apart from the money that actually moved. After that the entry only grows downward, by dated lines.

A log of decisions is useful only while its entries are contemporaneousSet down as an event happens, not assembled later out of recollection.. A record that always turns out to have been right is impossible to tell apart from a record composed yesterday. The moment an entry can be revised once the period has run, it stops proving anything. Every rule further down, about when an entry closes, about what may be added to it, about how a blank cell is marked, exists to protect that one property and nothing else.

One limit runs through everything the log can hold, and the limit is better said now than discovered later. The reading on the floating benchmark is known for the opening period and for that period only. Pricing the arrangement today needs a dated list of the readings the benchmark is expected to show at each payment still ahead, and no such list was assembled for these notes. The arithmetic therefore stops at the end of period one. The value column on the entry stays blank from top to bottom, and blank is its honest state rather than an oversight somebody forgot to correct.

One side of this arrangement sends the same amount every period and takes back whatever the benchmark happens to say. The fixed payer is Chitrakoot Cements Limited, a cement maker written for these notes, and the rate set into its half of the agreement is 7.20 per cent a year. Facing it stands Saranga Capital Limited, invented in the same way. Saranga Capital's half runs the opposite way round: what it sends out depends on a reading taken from the floating benchmark, and what comes back to it is the fixed 7.20 per cent. Both halves are sized against one shared figure, a notional of Rs 1,000 crore. The notional works as a multiplier and never itself changes hands.

Why keep a log at all, when the arrangement is already written down somewhere?

A cloth shop on a market street signs a rent at a fixed monthly figure for three years. On the day it is signed the owner writes four lines inside the cover of her ledger: what the landlord offered, what the street was charging that month, why a fixed figure suited a shop whose sales swing with the wedding season, and the one development that would make her wish she had stayed on a monthly deal. Two years on, the street rate has dropped. The four lines inside the ledger cover are now the only object in the shop that can tell her whether she chose badly or simply sat on the wrong side of a market. Without them she has a rent and a feeling.

The same gap opens around any arrangement of the kind this material has been working through. The confirmation says what was agreed. The schedule says when money moves. The ledger says what moved. Not one of those three holds a reason, and a reason is exactly what somebody will want a year later, when a period has run against the party and a question is asked about why the arrangement was entered into at all. Memory will not supply it either. Memory rewrites itself quietly in the direction of whatever happened. A decision log exists so that a decision can be reviewed on what was known at the time, and there is no other object in the house that stores that.

The log therefore rests on a single property. An entry is written before the outcome is known. Writing before the outcome is known is not a convention, a house style or a nicety; it is the whole of the log's claim on anybody's attention. An entry composed on the decision day carries a mixture of fact and belief that will never be available again in that combination. The moment the first reading arrives, the belief half starts being edited by knowledge. Preserved, that mixture is something no system of record can reproduce. Lost, it leaves the log a slightly awkward duplicate of the confirmation.

One rule follows from the property immediately, and it is the rule most often broken, almost always by somebody trying to be helpful. An entry is never edited afterwards. An entry is only appended to. Corrections, second thoughts, better phrasings, findings from a review: every one of them goes underneath, on a line carrying its own date. Nothing already written moves, softens or disappears. Ordinary good practice says to keep a document current, so the rule feels wrong the first few times. A decision log is the one document in the house where keeping it current destroys it.

Every entry also carries two figure columns that are held apart and never combined. One holds the figure the rates are worked on: for this arrangement, a notional of Rs 1,000 crore. The other holds the Rs 12.00 crore of difference that the first period actually settled at. Both are printed, both belong on the entry, and no line anywhere adds one to the other. The reason has been the argument of this whole material: a multiplier and a payment are not two amounts of the same kind, so their sum describes nothing that exists.

The log is where this sequence stops. It hands its output to nobody. Everything earlier in this material fed into something later. An entry in this log feeds only the person who opens it next, who may well be the writer twelve months older. Whatever a reader does with that entry belongs to them.

When is an entry opened, and when is it closed?

An entry is opened on the day the decision is taken and closed on the same day. Not the day the confirmation arrives, not the day the papers are filed, not the following Monday when there is time. The decision day, both ends of it.

An entry left open quietly acquires sentences written after the first reading was observed, and nobody reading it later can tell which sentences those were. Closing an entry on the day it was opened matters more than it sounds. Three weeks is more than enough. The writer means only to tidy up a half-finished thought, adds a clause, sharpens a sentence, and the entry now contains a mixture of what was believed on the decision day and what was learned in the fortnight after it, indistinguishable on the sheet. No reviewer can separate them. Neither, after a while, can the writer.

The practical form of the rule is short, and shortness is precisely why it slips. The entry carries its date. The entry also carries the name of the person who wrote it, so a later question has somebody to go to and responsibility does not float loose. The name is the entry's sign-offThe named person who accepted a piece of work, recorded so that responsibility does not float free of anybody., and it is a name rather than a department. The entry is closed when that person stops writing. Everything after that moment is an appended line with a date of its own, and the date is what makes it separable.

Consider what this does to the shape of the object on a calendar. The entry itself occupies a single day, however much is written in it. The additions occupy the months after it, scattered wherever a period settles or a review reports. Drawn out, the entry is a narrow block and the appended lines are marks arriving later along the same line. The shape is not a metaphor for the discipline; it is the discipline.

Which columns does an entry carry, and which two never merge?

An entry has four groups of cells. The first two are figure columns, held apart. The third is reasoning. The fourth holds the cells nobody can fill, and it has a block of its own further down.

The first column is what the agreement says, copied across and not interpreted. The two sides, named in full. The notional. The fixed rate with its period attached. A rate without a period is not a rate. Where the floating leg takes its reading from, and on which day. The counting method, recorded separately for each leg. The two legs need not share one. And the dates: trade, effective, reset, payment. Each of those has been worked through elsewhere in these notes, and each is written down here rather than taught here. The distinction between writing a term down and teaching it matters more than it looks: a decision log that starts explaining its own terms has begun turning into a textbook, and textbooks do not get kept up.

The second column is what the first period computes to, worked from the first column and nothing else. The fixed leg is built out of three things and nothing more: the notional, the rate written into the agreement, and a whole period of counting. Put those three together and it comes out at Rs 72.00 crore gross. Its opposite number is built the same way, from a reading of 6.00 per cent rather than the agreed 7.20, and comes out at Rs 60.00 crore gross. Subtract, and Rs 12.00 crore remains, leaving Chitrakoot Cements. Because the two legs fall due together and in the same money, only their difference actually travels.

There is a second route to that same difference, and it is worth writing into the entry because it checks the first. The two rates stand 1.20 percentage points apart. Applied to the same notional, that gap gives Rs 12.00 crore directly, without either leg being computed at all. The two routes agree exactly rather than approximately, and the arithmetic forces that agreement rather than a happy accident producing it: subtracting the legs and subtracting the rates before multiplying are the same operation in a different order. An entry that carries both is an entry somebody else can check in ten seconds.

Now the discipline that the whole of this material has been building toward. The two figure columns are kept separate and are never totalled together. Rs 1,000 crore is what the rates are worked on. Rs 12.00 crore is what actually moved. A line adding them would produce Rs 1,012.00 crore, a figure that corresponds to nothing anybody could point at. Print the two beside each other, put a rule between them, and leave the space where a total would sit empty. Twelve crore against a thousand crore comes to three parts in every two hundred and fifty, or 1.2 per cent written as a percentage, and that ratio is the honest way to relate the two. The sum is not.

Printed out as it would actually be kept, the entry looks like this. Two panels, a rule between them, and no total line anywhere on the sheet.

The third column is the reasoning, and it is the only part of the entry a machine could not produce. The arrangement is meant to do something for the party, and the column says what, in the writer's own language rather than in a form of words borrowed from a policy document. The development that would have to happen for the answer to stop being true. Ragged prose is fine here. Personal prose is better than tidy prose. Tidy prose is usually somebody else's.

Inside that column sits one line that a later reviewer will value above everything else on the sheet, and it is the line almost nobody writes at the time. What would change my mind. The line is short, it is uncomfortable, and it converts the reasoning from decoration into something that can be tested. A year on, a reviewer can hold that line against what actually arrived and ask a real question: did the named development happen, and if it did, was anything done? Without it, the reasoning column can only ever be read approvingly or sceptically, never checked.

How is a cell nobody can fill recorded?

Three cells on this entry hold no value, and the temptation with all three is the same: leave them blank and move on. Doing that costs more than it saves. A blank cell is read as an oversight by everybody who meets it afterwards, including the person who left it. Six months later nobody can tell whether the cell was considered and left alone or simply missed.

The first of the three is headed current value: the worth of the arrangement today. Filling it takes a dated list of the readings the benchmark is expected to show at each payment still to come, published by whoever publishes such things, and no such list stands behind these notes. So the cell carries that sentence instead of a figure. The sentence is not a gap in the entry; it is a statement about what the entry can and cannot support. A number produced without the list would be a fabrication a reader had no way of detecting, so the sentence is more useful than a number would have been.

The other two cells are empty for a completely different reason, and treating all three alike is exactly the mistake to avoid. One asks how long the papers behind this arrangement stay with the party who kept them. The other asks which details of it travel onward, to whom, and inside what window. Both of those have answers. The answers live with the Reserve Bank of India, at rbi.org.in, and they shift as time passes, so a figure typed into either cell would be quietly untrue by some date nobody had printed. The cell therefore carries the routing rather than the answer.

Which gives the discipline that covers all three, and it is the one habit from this guide most worth carrying into any record at all. A cell carries either a value or the reason it has none. Never a blank, and never a dash. A dash is worse than a blank. A dash looks deliberate without saying anything.

What happens to an entry once a period settles?

A period runs, a reading is observed, an amount is computed, both sides agree it, and money moves. All of that is information the entry ought to carry. None of it is allowed to touch a word already written.

A settled period arrives as a dated line appended beneath the entry, carrying four things: the reading that was observed, the amount computed from it, the amount agreed with the other side, and the amount actually paid. Four, not one. The four are separable, and the gaps between them are where the interesting problems live. A computed amount that differs from an agreed amount is a conversation with a counterparty. An agreed amount that differs from a paid amount is a conversation with an operations desk. Collapse the four into one figure and both conversations become invisible. The line carries its own timestampThe date, and sometimes the hour, fastened to a line so its place in a sequence cannot be argued with., and the date is what keeps it separable from everything written on the decision day.

Now the part that requires actual restraint. The reasoning column is not revised in the light of what happened, even when the writer can now see that it was poorly expressed. The temptation here is real rather than theoretical. The urge is not to hide anything; it is to be clearer. A sentence that read fine in the moment reads naive after a period that ran the other way. The writer wants to fix it. Fixing it removes the only property that made the column worth keeping: the column is what they thought then, in the words they had then. Leave it. If the thought was badly put, note that on an appended line, dated, and let both versions stand.

There is a second kind of appended line, and it is where the log joins the periodic review, covered separately. A review reads the log, works through what was decided and how, and produces findings. A finding does not go into a report that lives somewhere else and is never opened again. A finding arrives as a dated line hung on the step of the entry it concerns, dated to the review rather than to the decision. Dating the line that way lets the two routines join without either one overwriting the other: the review gets its evidence from an entry it cannot alter, and the entry gets its correction without losing what it originally said.

The uncomfortable part is that tidying does not feel like a restatementA figure or a sentence issued again in corrected form, replacing whatever stood there before it. while it is being done. A restatement announces itself; tidying feels like housekeeping. The writer intended nothing, so intention is not the test. The test is whether a reader, holding the entry, can place its sentences in time. After a rewrite they cannot, and neither can the writer once a few months have passed.

Which mistaken version of a log is somebody's real practice?

Three mistaken versions of this object are somebody's real practice somewhere, so each is worth naming plainly rather than implying.

A decision log is not a diary. A diary has an entry for a day; a log has an entry for a decision. Nothing is written on a day when nothing was decided, and a week with four decisions in it carries four entries. Run it as a diary and two things go wrong at once: entries appear that record nothing, and decisions taken on busy days get folded into a single paragraph with a shared date. Folding them destroys the very separability the dating was for.

The log is not the system of record for the arrangement. The confirmation, the schedules, the amendments and the settlement history live in a system built to hold them, with controls this log does not have and does not need. An entry copies across the handful of terms it requires so the reasoning has something to sit against, and duplicates nothing beyond those. A log that grows into a shadow copy of the confirmation acquires a second problem immediately: the two versions can now disagree.

The log is not a report to anybody, and this third mistake is the quietest failure in the whole of record keeping. An entry written with a reader in mind stops recording what the writer actually thought and starts recording the reasons the writer is prepared to have read. The reasons become defensible rather than real. Nothing about the log looks any different, and that is the problem: it will pass every inspection it was written for and answer no question anybody genuinely has afterwards. The closest useful comparison is with an accountant's working papersThe rough sheets somebody keeps while doing a piece of work, held apart from the finished output., kept rough on purpose. Polish is where the useful detail goes to die.

Who actually opens an entry they did not write?

The value of this discipline shows up entirely in the hands of people who were not in the room. Four of them, and each opens the entry looking for something different.

The successor. Somebody inherits a book of arrangements when the person who put them on leaves. Without a log they inherit positions and no reasons, and every one of those arrangements then has to be re-justified from scratch or carried on faith. With a log, the first thing they read is the reasoning column and the line saying what would change the writer's mind, and within an afternoon they know which arrangements still answer a live need and which are running on inertia.

The credit officer at a lending bank. The credit officer is not judging the arrangement; the officer is judging whether the borrower runs a house where decisions are taken deliberately. A log that shows terms, arithmetic, a stated reason and a stated trigger, entry after entry, tells them something about the borrower that no single financial statement does. A log full of tidy defensible sentences tells them the opposite, and experienced readers spot the difference quickly. Real reasons are uneven and manufactured ones are uniform.

The internal reviewer. Internal reviewers rarely read every entry. The reviewer pulls a handful, an exercise in samplingChecking a handful of items drawn out of many, then reasoning about the whole set from what the handful showed., and reasons about the log from what the handful shows. Sampling is why a single tidied entry does damage out of proportion to itself: once a reviewer knows entries can be revised, the sample stops telling them anything about the population, and the only honest conclusion available is that the log cannot be relied upon.

And the writer, twelve months older. The writer is the reader most people underestimate, and the one who benefits most. A household running on a single salary knows the feeling exactly: the decision to fix the school fee instalments rather than pay as they fall due made obvious sense in the month it was taken, and by the following year nobody in the house can reconstruct why, only whether it turned out convenient. Four lines written at the time settle the argument. The person best served by an entry written honestly is almost always the person who wrote it.

What shows whether the log is any good?

Everything above compresses into one question, and it is a question a keeper can run on their own log this afternoon without asking anybody's permission. Could somebody who was not in the room review this decision from the entry alone?

Passing it requires a short list, and short is the point: a test with fourteen conditions is a test nobody runs. The terms are present. The arithmetic can be redone from what is written, so the notional, the rate, the reading and the counting method are all there rather than three of the four. The reasoning is in the writer's own words rather than in a form of words. Each unknown is recorded as unknown, carrying what would resolve it. And every later addition carries its own date.

One thing has to be said before that test gets used on anybody's real log, and this material holds it without softening. An entry recording a period that ran against the party is not a record of a failure. A period runs the way it runs. Ending up facing the direction it took is nobody's failing, and a log kept as though it were will be quietly abandoned inside a year. Nobody writes honestly into a document that will be used against them. Abandonment is not a soft point about morale; it is the practical condition for the log continuing to exist at all.

What does a complete entry still not give?

A complete entry lets a decision be reviewed on what was known when it was taken. Review on what was known is a real thing, and it is worth the twenty minutes an entry costs. A complete entry is also all of it, and the three things an entry does not deliver are worth naming so nobody waits for them.

Should either side be in an arrangement like this? Record keeping is not where that gets settled, and these notes settle it nowhere else either. Four things stand between anybody and an answer, and not one of them is settled here. How the party borrows. When its cash actually arrives. Whether it can sit through a period that runs the other way. And where the benchmark goes next, the very thing that must never be written into an entry as though it were settled.

A second absence is the worth of the arrangement, on the decision date or on any date since. The value column stays empty, and by now the reason should be familiar rather than surprising: no dated list of expected benchmark readings stands behind these notes, and any figure printed in that column would be resting on a list nobody has, with none of that shakiness showing in the entry. The emptiness is stated as a sentence rather than as a symbol. A marker whose meaning the reader invented is exactly what gets filled in with a guess six months later.

And the log says nothing about what the next reading will be. The silence is not a shortcoming; it is the reason the reasoning column exists in the first place. Nobody writes down what would change their mind about a thing they already know. A counterfactualA statement about what would have followed had a choice gone the other way instead. written a year later, about what the party would have done had the benchmark gone elsewhere, is worth nothing at all unless the original entry fixed what the party actually expected. Chitrakoot Cements is the fixed payer up to the final payment date and not one period beyond it, and every period in between will be recorded the same way, appended and dated.

A log is a thing that gets written, not a pair of quantities one of which chases the other, and the single relationship an entry does hold, between the figure the rates work on and the money that travels, is covered separately.

Keeping this log is a habit of writing. An entry is not a signal to act on, and an entry that reads like one has been written for the wrong reader.

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