How to Run a Derivatives Decision Post-Mortem on Process

A decision post mortem asks whether a decision was reached well, using only what could be known on the day it was reached. A post mortem is not a hunt for whoever was wrong. The review stands at the decision date, rebuilds what the decision maker had in front of them, keeps the process question and the outcome question in separate columns, and ends in a finding attached to a step.

Underneath that sits a separation which sounds obvious the moment somebody says it out loud, and which almost nobody holds in practice. How a decision was made and how the period turned out are two different questions, and answering the second does not answer the first. A careful decision can be followed by a stretch that runs against the party who took it. A sloppy one can be followed by a stretch that runs in its favour. Nobody in the room chose the reading at a reset, so neither of those says anything about the reasoning.

Ordinary life carries the same lesson. Somebody carries an umbrella on a morning the forecast called wet, and the day stays dry. The umbrella carrier was not foolish; they were unlucky in a way that cost them nothing. Somebody else leaves the umbrella behind on the same morning and also stays dry, and nobody calls that judgement. Judging both by the weather makes the second look clever. Judging both by what was on the forecast makes them what they actually were. Everything below is a way of holding on to that distinction when there is money on the table and somebody is upset.

A decision can be looked at from two positions, and it helps to have names for them. Judging ex anteLooking forward from a point in time, using only what was on the table then. The opposite view is the one taken once the result is in. means standing at the moment of choice and asking what the chooser had. Judging ex postLooking back once the result has arrived. Good for describing what happened, useless for deciding what somebody could reasonably have known beforehand. means standing after the result and reading backwards. A post mortem is an ex ante judgement conducted at an ex post moment, and every move below exists to protect that awkward combination from collapsing.

Who are the two sides here, and which questions sit outside the review?

Chitrakoot Cements Limited, an invented manufacturer, took the side of the agreement that hands over a rate settled in advance. Sitting opposite is Saranga Capital Limited, an invented lender, whose side hands back whatever the floating benchmark happens to read. Every figure below belongs to that one arrangement, so a review can be run on something concrete instead of on an abstraction.

One caution before the routine starts, and it shapes everything that follows. Putting a value on the arrangement takes a schedule carrying, date by date, where the floating benchmark is projected to stand. No party ever assembled such a schedule, so the review runs on the single period the record does describe, and wherever a value would sit, the reason for the blank is written in instead. The absence carries the central claim of the routine: a complete and useful review can be run without the ending being in the room at all.

What is a review like this actually for?

A review like this is for finding out whether the way a decision was reached would produce a good decision again. The purpose stops there. A routine that answers that question is worth running every period; a routine that answers any other question is worth running once.

The thing it gets used for instead is establishing who was wrong. The blame version is easier to run and more satisfying in the room, and it stops producing findings inside about a week, once the people whose written reasoning is being read learn what it is being read for. Somebody has to record the terms, so the terms keep getting written down. Nobody has to record the reasoning, so the reasoning stops. Two periods later the reviewer opens the file and there is nothing in it to review.

There is a second failure that arrives before the reviewer has asked a single question, and it is about size. The agreement records a notional of Rs 1,000 crore. The notional is not a commitment of Rs 1,000 crore but a multiplier, so a review that opens by asking how a commitment of that size was ever approved has begun on the wrong quantity. Rs 12.00 crore moved across the first period, and the notional sat there being multiplied by, exactly as it was written to do.

Before any move is made, it helps to see the shape of the finished thing. A completed review has two columns that never answer each other, a rebuilt sheet dated to the decision, a short set of questions that can actually be answered, at least one cell drawn empty rather than guessed at, and a finding attached to a step rather than to a person. Six moves lead there, and the first one decides everything after it.

Move one: which date does the review stand at?

The decision date. Not today, not the payment date, not the day somebody noticed the money had gone the other way. Everything the reviewer is permitted to use is what existed on the day the decision was taken, and everything dated after it is out of bounds as evidence about the reasoning.

Fixing the date comes first because it is the only move that cannot be recovered later. Once a reviewer has seen how the period settled, they cannot unsee it. Every judgement made afterwards is being made by somebody who knows the answer and sincerely believes they are setting it aside. The reviewer is not setting it aside, and there is no way to check whether they are, including by asking them. The knowledge does not sit in a compartment; it changes which steps look worth examining and which look like nothing much.

Which means stating the rule does not achieve it. The rule needs a practical form, and the practical form is a piece of paper. Everything known on the decision date goes down in writing before anything dated after it is opened. The terms as they stood. The reading that had been observed. The working that had been done. Whatever reasoning was recorded at the time. The sheet is then closed and treated as the whole of the evidence for the process column. Something dated later can then be filed where it belongs, rather than seeping into a judgement about reasoning.

Move two: why must the two columns never answer each other?

Draw two columns. One asks whether the process was sound. The other records what happened afterwards. The two columns sit side by side, they get filled in independently, and neither is ever used to settle the other.

The reason for drawing them rather than merely intending them is that the separation is unstable in the head and stable on paper. Left in the head it collapses under the first strong result, in either direction. Set out as two columns the separation survives. A reviewer writing in one column can see that the other one is a different column.

Seeing all four combinations at once is what makes the separation stick, so the columns are best drawn as a grid. Sound process with a period that ran in the party's favour. Sound process with a period that ran against. Careless process with a period that ran in the party's favour. Careless process with a period that ran against. All four happen, all four are ordinary, and the reason the separation feels strange is that only one of the four ever gets a review called on it.

Which leaves one sentence a reviewer is not permitted to write, and it is the whole failure compressed into a line: the period settled against the party, therefore the decision was wrong. Writing that sentence lets the second column answer the first. Preventing it is the only thing the grid exists for, and the sentence turns up in almost every review that has not been built to keep it out.

Move three: what does the reviewer rebuild, and out of what?

Out of the record, and never out of anybody's memory. Four things get pulled onto one sheet: the terms written into the agreement, what was observed on the dates that mattered, what was computed from the two of them, and what reasoning was set down at the time. Three of those four can be pulled by anybody with access to the file. The fourth cannot be pulled at all unless somebody once wrote it.

The first three are mechanical. A reconstruction that guesses at a source is not a reconstruction, so precision about where each number is found matters more at this point than what the number means. The terms come off the agreement itself. The notional sits in the schedule. The rate on the fixed side sits beside it. The counting method sits in the definitions. The observed reading comes off whatever the party keeps as its record of readings, carrying the date it was observed on. The working comes off the settlement calculation, and if no working was kept, that is already a finding.

The fourth is different in kind. Reasoning written at the time is a contemporaneousSet down while the event was still going on, instead of reconstructed later out of memory. record, and it either exists or it does not. More often than not, nobody wrote it. A decision whose reasoning was never written cannot be reviewed on process at all, so where the fourth box comes up bare, the reviewer records the blank as a finding in its own right rather than working around it. The blank is not a technicality. There is no substitute for it. Recollection assembled after the result is known arrives already shaped by the result. Shaping of that kind is precisely what the first move was built to keep out, and no amount of good faith removes it. All three of them know how it ended, so asking three people and taking the average does not corroborateTo back a statement with a second, independent source, so that it does not rest on one person saying so. anything either.

How long a party has to keep its paperwork on a privately agreed arrangement, and which pieces of that paperwork it has to keep, is settled by the Reserve Bank of India at rbi.org.in. The review works from whatever the party actually holds rather than from what it ought to hold. Where the two differ, that gap is worth a finding too, and it is a finding about the records rather than about anybody's judgement.

Move four: which questions can actually be answered?

Only questions about the process, and there are fewer of them than people expect. Four carry most of the weight.

Was the obligation on each side described correctly at the time, plainly enough that whoever read the description knew what was owed and by whom. Was the arithmetic shown rather than asserted, in a form anybody picking up the sheet could have checked and found the same answer from. Were the things nobody knew recorded as unknown, in the body of whatever was written rather than tucked beneath it where a reader skimming the sheet slides straight past them. And was the reason written down before the period ran, rather than after it. The other three questions depend on that one.

All four share the test that lets a reviewer build a list of their own: each can be answered from the reconstructed sheet alone, with no reference at all to how anything turned out. The rule stops there. A candidate question is held up against the rebuilt sheet, and the test is whether the sheet settles it. If it does, the question belongs in the process column. If settling it needs the ending, the question is about the world rather than about the decision, and it goes in the other column or nowhere.

Which disposes of one question asked more often than any other and belonging least: did the benchmark move the way anybody thought it would. The sheet predates the movement, so the sheet cannot answer it. The movement is a fact about what the world did, and the world was not consulting anybody's reasoning when it did it.

Move five: which cell cannot be filled at all?

Every review of an arrangement like this reaches for one figure early: what was the thing worth at the review date. Here that figure is unavailable, and the interesting part is what a reviewer does about it.

Such a figure is produced by a set of expectations for the floating benchmark, projected forward and dated, one for each period still to come. No party ever assembled such a set for this arrangement. So the row is drawn, the reason is written inside it, the source that would close it is named, and the review moves on to everything that is actually knowable. A party that cannot say what an arrangement is worth at a review date has learned something about its own records, so the gap gets recorded as a finding, and that finding is about the records rather than about the decision.

A second row stays empty for a completely different reason, and telling the two apart is worth a moment. Whether the arrangement qualifies for hedging treatment, and what evidence has to be on file before it does, is settled by the Reserve Bank of India at rbi.org.in. The hedging row is not blank because somebody failed to keep something. The row is blank because the answer belongs to an authority, the answer gets revised, and a row filled in from memory would carry an answer that was flatly incorrect rather than merely stale.

Move six: what does the review produce, and where does it go?

A finding. One sentence, attached to a step. The step might be a station in the period, a part of what was written down, or a field in the record that gets kept. The test of a finding is that it points at something a person can change on a Tuesday afternoon.

Two things a finding is not, and this is where reviews collapse into theatre. A finding is not a verdict attached to a person. A verdict travels to an appraisal rather than to a step, and once it travels there the written reasoning stops. Nor is a finding a general instruction to take more care. Nobody has ever taken more care as a result of being told to. Care is not the variable. The variable is what gets written down.

A review whose output has nowhere to go will not be run twice, so a finding also needs somewhere to land. Where it lands is the next entry of the decision record: the finding changes what that entry is obliged to contain. A separate note takes up how that record is kept and what a good entry looks like. For the purposes of this routine the only thing that matters is that the finding travels, and that a reviewer can name the entry it travelled into.

One more rule, stated out loud rather than assumed. Landing on the far side of a reading is not a failure and must never be written up as one. The reading went where it went, and nobody in the room chose it that morning. A review drafted as though somebody had slipped up will be accurate about nothing, it will be the last honest one anybody files, and the cost of that lands on whoever runs the review after next.

What does the whole review look like, run on one arrangement?

The six moves are worked end to end below on the arrangement between the two invented companies, for the first period only. No result of any kind was ever recorded for this arrangement, so the review runs entirely in the process column, and the outcome column says so where a result would sit.

Move one fixes the date. The review stands at the day the agreement was entered. Everything after it is available as description and unavailable as evidence about the reasoning.

Move two draws the columns. Two of them. The process column gets filled. The outcome column gets its heading and then a sentence explaining why nothing sits beneath it. How the period actually ended is simply not in this material. Neither is any history of the periods before it, any statement of how likely a given reading was, nor any spread across the readings it might have taken, so there is nothing honest to write there. Inventing something would produce a figure a reader could not distinguish from a real one. A blank is the better outcome.

Move three rebuilds the sheet. The terms come off the agreement: a notional of Rs 1,000 crore, the fixed side set at 7.20 per cent for a year, and a counting fraction of 1.0000, the figure a full period without any awkward part month works out to. The observation comes off the reading record: the floating benchmark stood at 6.00 per cent a year for this period. The computation follows from the two.

Both sides fall due together and are counted in the same money, so nobody wires Rs 72.00 crore in one direction while Rs 60.00 crore travels the other way. The pair collapses to a single movement of Rs 12.00 crore, from Chitrakoot Cements towards the other side of the agreement. There is a second route to that figure which arrives at exactly the same place: the two rates stand 1.20 percentage points apart, and applying 1.20 percentage points to Rs 1,000 crore also produces Rs 12.00 crore. The two routes agree because they are the same operation with the notional factored out. The agreement is forced arithmetic rather than a check.

Move four asks the four questions. Was the obligation on each side described correctly at the time: on this record, yes. Both sides are named and each one's leg is stated. Was the arithmetic shown rather than asserted: yes, and the table above shows it in the form the answer should take. Were the unknowns recorded as unknown: partly. The absence of any expectation for later periods is stated above but was not stated in the original file. Was the reason written down before the period ran: no, and that is the finding.

Move five draws the two empty rows. The row on what the arrangement was worth at the review date, with the missing schedule named inside it. The row on hedging treatment, with the Reserve Bank of India named inside it at rbi.org.in.

Move six writes the findings. Two of them, each attached to something rather than to somebody. The first: no reasoning was recorded before the period ran, so the next entry in the record has to carry a short written reason at the point the terms are agreed, not afterwards. The second: the party cannot produce a dated set of expected readings, so the next entry has to name the source it would come from and say plainly that it is unavailable. Neither finding mentions a person. Both change a document.

What would go wrong if a reader could drag the reading?

The pair a control would naturally animate here is a decision set against an outcome, and no outcome exists anywhere in this material. A control that slid one along would mean inventing exactly that, and a reader could not tell the invention from a record.

Worse than that, it would teach the reverse of the lesson: the entire routine exists to stop a review turning on which way a reading went. A dial that made the reading movable would put the reading back at the centre of the exercise, where it sat before anybody drew two columns. The grid of four combinations does the work a control would have been reached for, and it does it without putting a number behind any of the four cells.

Who actually runs one of these, and what do they get out of it?

The shape is the same and the money is smaller, so the household version comes first. A couple takes a home loan at a fixed rate. Two years later, floating rates have fallen and the neighbour who chose floating is paying less. The kitchen table review is short and it goes: the wrong one was chosen. But that is a review of the rates, not of the choosing. The useful review asks a different set of things. Did anybody write down why fixed was picked. Was it because the household could not absorb a rise in the instalment on one salary, a reason that is still true. Was the arithmetic on the instalment actually done, or estimated. Were the things nobody could know, such as where rates would go, written down as unknown, or quietly treated as known. The second version of the conversation produces something the household can use next time. The other version produces an argument.

A treasurer at a company running an arrangement like the one above does the same thing on a bigger sheet. Every period they hold a short session, half an hour at most, working the six moves against whatever was decided that period. The output is one or two findings, and the findings almost always concern a document rather than a judgement: the reason was not recorded before the terms were agreed, or the observation record does not carry the date the reading was taken, or nobody can produce the working that turned the reading into the amount that settled. Each of those is fixable by changing what a form asks for.

A lender looking at the same company reads the reviews rather than running them, and what they are reading for is whether a routine exists at all. A borrower who can produce a written reason dated before each period, and a record of findings that changed something, is a borrower whose arrangements are being run rather than merely held. A borrower whose file has terms and settlements and nothing else may be perfectly sound and cannot demonstrate it. The difference between the two files shows up long before any difficulty does.

An analyst covering the company uses the same reading in a narrower way, and here it pays to be careful about what is being claimed. A review speaks to process and not to results, so it does not support any attributionSplitting a result into the separate causes that produced it, so that each part can be examined on its own. of a company's numbers to its arrangements. A review supports a description and nothing more: this party writes reasons down beforehand, or it does not. Anything stronger needs data this material does not contain.

One practical detail that decides whether any of this survives contact with a real office. The person who runs the review should not be the person who holds the sign-offThe recorded approval that lets something proceed, naming who agreed to it and when. on the decision being reviewed. Not because anybody is suspected of anything, but because the reviewer needs to be able to write down that the reasoning box was empty without that sentence reading as a confession.

What does a finished review not establish?

A completed review says whether a decision was reached in a way that could be defended on the day it was reached, and it says what to change about the record before the next one gets taken. The pair of answers is genuinely useful and is not the pair most people were hoping for.

A review does not say whether the party ought to have entered the arrangement at all. Should Chitrakoot Cements have signed? Answering that would take a view on what a particular reader's own position needs, and no general routine can hold that view. The routine can say which things somebody would have to know first, and that list is longer than it looks.

A review does not say whether the arrangement was a good one. Judging that needs what the arrangement was worth at a series of dates, and no such series sits in the record. Nor can a review say what the next period will do. Nobody can, and that is the reason the process column exists in the first place: if the future were readable, judging decisions by results would work, and there would be no need for any of this.

One more limit, and it is the quietest. A single review is a sample of oneA single instance treated as though it showed a pattern. One case can illustrate something; it cannot establish how often anything happens.. A single review can show that a particular decision was reached well or badly. A single review cannot show that the party generally reaches decisions well or badly, and a reviewer who writes the second conclusion from the first has made a smaller version of the mistake the whole routine is built to avoid.

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