Sensitivity Measures: Moving One Variable at a Time

A sensitivity measure moves one market variable, holds everything else still, and reports the change in the value of the position. At Vindhya Commercial Bank Limited, invented, one basis point moves rate sensitive assets by Rs 25.2 crore and rate sensitive liabilities by Rs 21.0 crore, so the value of equity moves by Rs 4.2 crore a basis point. Every duration here is that bank's own locked input.

Read that answer again and notice how little it claims. A sensitivity measure does not say what rates are going to do. It does not say how likely a move is. Nor does it say whether Rs 4.2 crore a basis point is a lot or a little for a balance sheet of this size. A sensitivity measure is a conditional statement and nothing more: if this one thing moves by this much, the position is worth that much more or less. Almost every argument about one of these numbers turns out, on inspection, to be an argument about the condition rather than about the arithmetic.

What is a sensitivity measure actually reporting?

A sensitivity measureA number produced by moving one market variable by a stated amount, holding everything else still, and reporting the change in value. is built out of three decisions and one multiplication. The three decisions are: which variable is moved, how far it is moved, and what is held still while it moves. The multiplication is the easy part. The three decisions carry all the meaning, and they are the part almost nobody writes down where the number appears.

The shape is easier to feel away from banking, on something concrete. The owner of a small tea stall wants to know how exposed the stall is to the price of milk. One option is a very large question: what happens to the stall if the economy turns? The large question has no answer anybody can act on. The other is a very small one: if milk goes up by Rs 1/- a litre, and the stall sells exactly what it sold yesterday, and none of its prices change, what happens to today's takings? At forty litres a day, the answer is Rs 40/- a day, and it arrives in one line. The answer does not reveal what milk will do. The answer reveals what a rupee of milk is worth to the stall, and that is a completely different and far more useful thing to know.

The tea stall calculation is a sensitivity measure. The variable was chosen, and it was the milk price. The size of the move was chosen, and it was Rs 1/- a litre. Everything else was held still, and saying that the same volume would sell and no prices would change is how the holding still got announced. And the number that came out, Rs 40/-, is a statement about the stall rather than a statement about the milk market.

Now put a bank in the tea stall's place. Vindhya Commercial Bank Limited, invented, carries Rs 84,000 crore of rate sensitiveA position whose value or income changes when the general level of rates changes, whether or not it is traded. assets and Rs 84,000 crore of rate sensitive liabilities on a balance sheet of Rs 96,000 crore. The variable it chooses to move is the general level of interest rates. The size of the move it chooses is one basis pointOne hundredth of one percentage point.. One basis point is one hundredth of one percentage point. The bank holds everything else still: which currencies it is exposed to, the spread between one bond and another bond, the shape of the rate curve, how much anybody has drawn on a committed line, and every balance on the sheet. Then it reports what the value of equity is worth after that move and only that move.

The reason to be pedantic about the third decision is that it is the one that gets lost first. A number that reports what happens when one thing moves has, by construction, assumed that nothing else moved. The assumption is not a flaw hidden inside the measure. The assumption is the measure. If it were not there the number would not be computable, and a reader who forgets it is there will read the number as covering ground it never touched. Vindhya Commercial Bank Limited has a worked case, set out below, where forgetting the assumption cost real money.

Why does the measure move only one variable, and what does that choice cost?

Because a number that moves two things at once cannot show which of them did the damage, and the whole purpose of putting a figure in front of a committee is so that somebody can decide which lever to pull.

The tea stall again, but this time with three things moving together: milk goes up, sugar goes up, and the electricity board raises the tariff on the fridge. Takings fall by Rs 220/- a day. Acting on that is impossible. Renegotiate with the dairy, switch to a cheaper sugar, or replace the fridge? The single number of Rs 220/- has no answer in it. The answer was destroyed at the moment the three moves were combined. Split into three separate one at a time measures, it becomes Rs 40/-, Rs 25/- and Rs 155/-, and the fridge is now the thing to go and look at.

The tea stall split is the case for one at a timeThe design choice that makes a sensitivity measure readable and makes it blind to two variables moving apart., and it is a strong case. A one at a time measure is attributable. A hedge can be added against exactly the variable it names, the measure recomputed, and the number seen to fall. A one at a time measure can be handed to a named person as their number, together with the lever that moves it. Attribution is why sensitivity measures survive on every risk pack in the world despite everything that is wrong with them, and it is why this invented bank's four market risk limits, L5 to L8, are each written against something that a single named lever can move.

Now the cost, and it is not small. Everything that lives in the interaction between two variables is invisible to a measure that never moves two variables. Suppose milk and sugar always rise together because both are trucked on the same diesel. A stall owner who measures each one separately then has three numbers that are individually correct and a portfolio view that is wrong. He will never see a day where both move at once until that day arrives. The one at a time design buys attribution and pays for it with blindness to everything that moves together, and that is a trade rather than a defect.

The word blindness is precise. The measure is not approximately right about combined moves. The measure has no opinion about them at all. Nobody ever asked it the question. Breach B5 below is where this invented bank paid for exactly that, and in that breach the measure behaved perfectly while the reading of it did not.

What is basis point value, and why is it stated per basis point?

Basis point valueThe change in the value of a position for a one basis point move in the relevant rate, stated per basis point because it is a local statement. is the interest rate member of this class of measure, and it is the one most often met. Basis point value answers a single question: how many rupees does this position gain or lose if the relevant rate moves by one basis point? Nothing else. Not how likely that is. Not what happens at fifty basis points, except by an assumption stated explicitly below. One basis point, one answer, in rupees.

The multiplication has three inputs and no more. Take the size of the position in rupees. Multiply by its modified durationA locked input in this case that turns a rate move into a value change. It is used as a number here and is derived in the fixed income subject area.. Modified duration is the number that turns a rate move into a value move. Multiply by the size of the rate move expressed as a decimal. For one basis point that decimal is 0.0001. The three inputs are the whole calculation. Balance, times modified duration, times the move, gives the change in value, and there is no fourth term hiding anywhere.

Where does the modified duration come from? At this invented bank it comes from the bank's own book: a locked input of 3.00 years on the asset side and 2.50 years on the liability side, used rather than derived. How a duration is built out of the timing of a bond's cash flows is a proper subject in its own right and is covered under fixed income. Here it is a number that arrives, gets used, and is named as the bank's own.

Now the part of the name that confuses people: why per basis point, when nobody at this bank has ever run a scenario of one basis point? Its own internal scenario is a 200 basis point move. So why not quote the answer at 200 and be done?

Because basis point value is a local statementA result that is accurate for a small move and progressively less accurate for a large one, which is why a per basis point figure carries the scale it was measured at.. The relationship between a rate and a value is not exactly a straight line, and it gets less straight the further the journey runs along it. Over one basis point the departure from a straight line is too small to be worth arguing about. Over four hundred, it is not. Quoting the answer at the smallest useful unit is a way of being honest about that: the number is exactly what it says at one basis point, and everything beyond that is a scaling performed by somebody, on their own assumption of straightness, and that assumption is then theirs to defend.

The scaling is done below, to the bank's own 200 basis point scenario, and the assumption inside it is stated in advance: the scaling assumes a straight line that a real book does not exactly follow. Saying it once at the top is cheaper than defending it later at a meeting.

What is the basis point value of this whole balance sheet?

Now the build, in full, on this invented bank's own locked inputs and on nothing else. There are exactly two inputs on each side and there is no step that is not written down here.

The asset side first. Vindhya Commercial Bank Limited, invented, carries Rs 84,000 crore of rate sensitive assets at a modified duration of 3.00 years. Multiply: 84,000 times 3.00 times 0.0001 gives Rs 25.2 crore. Rs 25.2 crore is what one basis point is worth on the asset side of this balance sheet, and it is a fall in value when rates rise. A longer promise to receive money is worth less once money in general is more expensive.

The liability side next. The same Rs 84,000 crore of rate sensitive liabilities, at a modified duration of 2.50 years. Multiply: 84,000 times 2.50 times 0.0001 gives Rs 21.0 crore on the liability side. The Rs 21.0 crore figure is the liability side basis point value and nothing else; it happens to be numerically identical to the realised loss of Rs 21.0 crore recorded on backtesting exception X3 in this same case, and the two have nothing to do with each other, so the object is named every time the figure is printed.

Here is the step that trips people, and it is worth being slow about. The liability side also falls in value. But the bank owes those liabilities rather than holding them, so a fall in what they are worth is a gain to the bank, not a loss. When a party has promised to pay somebody Rs 84,000 crore over time and the worth of that promise drops, that party is better off. So the two effects run in opposite directions on the value of equity, and the net is a subtraction rather than an addition. Rs 25.2 crore of loss on the asset side against Rs 21.0 crore of gain on the liability side leaves Rs 4.2 crore a basis point reaching the value of equity, and that is the number this whole guide is about.

Now scale it to the bank's own scenario, and say out loud what the scaling assumes. Vindhya Commercial Bank Limited runs an internal parallel moveA rate change applied equally at every maturity, which is the shape this bank's own internal scenario uses and is a stated simplification. of 200 basis points. The 200 basis points is its own choice and not a requirement from anywhere. At 200 basis points the asset side moves Rs 25.2 crore times 200, being Rs 5,040 crore. The liability side moves Rs 21.0 crore times 200, being Rs 4,200 crore. The difference is Rs 840 crore, and that is the change in the value of equity under this bank's own scenario. Multiplying by 200 assumes the relationship stays straight over that distance, and a real book does not exactly deliver that.

Then the check, and it is worth doing every time one of these is built. The same institution can be measured a completely different way: the duration gap is 3.00 years on the assets less 2.50 years on the liabilities, being 0.50 years, and the shock applies to the whole balance: 0.50 times 2.0 per cent times Rs 84,000 crore gives Rs 840 crore. Same answer, arrived at from a different direction. Two routes landing on the identical Rs 840 crore is a proof that the arithmetic holds together, and it is not a second figure to be reported alongside the first. Against tier 1 capital of Rs 6,600 crore that Rs 840 crore is 12.7 per cent, and against the Rs 990 crore cap of limit L8 it is 84.8 per cent utilisation. Both of those percentages describe limit L8 at this invented bank and nothing else.

Why is the headline a small difference between two large numbers?

Look again at the three figures at 200 basis points: Rs 5,040 crore, Rs 4,200 crore, and Rs 840 crore. The reported number is the smallest of the three by a long way. The reported number is 16.7 per cent of the movement on the asset side alone. The shape is not an accident of this bank and it is not unusual. A bank funds long assets with slightly shorter liabilities and the two sides very nearly cancel, so a difference measure looks like this on almost any balance sheet.

The consequence is worth stating carefully. The statement sounds like a criticism and is not. When a reported figure is the difference between two much larger figures, an error in either of the large ones lands on the small one many times magnified. Suppose the true asset duration were 3.03 years rather than 3.00, a one per cent error that no committee would ever notice in the input. The asset side would move Rs 5,090.4 crore rather than Rs 5,040 crore, an increase of Rs 50.4 crore. The whole Rs 50.4 crore lands on the difference, taking the headline from Rs 840 crore to Rs 890.4 crore, a rise of six per cent. A one per cent slip in an input became a six per cent slip in the answer, and the arithmetic did nothing wrong.

Notice which input carries the weight. The two balances are identical at Rs 84,000 crore, so they cancel exactly and cannot move the answer at all except through the durations. On this balance sheet the answer is carried entirely by the difference between 3.00 years and 2.50 years, so every rupee of the headline rests on two estimates rather than on two counted balances. A balance can be counted. A modified duration is built from assumptions about when cash flows arrive, and one of the largest of those assumptions at this invented bank is how long the Rs 36,000 crore of current and savings balances behaves as though it lasts. A committee makes that judgement; no ledger records it.

The household version is the honest one. A couple works out their monthly surplus by taking income of Rs 95,000/- and subtracting outgoings of Rs 89,000/-, leaving Rs 6,000/-. Five per cent is nothing at all in household terms. If the outgoings estimate is wrong by that much, that is Rs 4,450/-, and the surplus has been wrong by nearly three quarters. Nobody would call the subtraction a bad method. The right response is to spend the effort on getting the outgoings right, and it is exactly the right response here too: put the scepticism into the two durations, not into the minus sign.

How is a sensitivity measure different from the position it describes?

Because one of them is the exposure and the other is a sentence about the exposure, and confusing the two is the commonest structural error a new reader makes on a risk pack.

An open positionThe exposure itself, unhedged, to a market variable. A sensitivity measure describes it; the two are not the same object. is a thing the institution is holding. At Vindhya Commercial Bank Limited, invented, there are five currency positions numbered FX1 to FX5: a long US dollar position of Rs 144 crore, a short euro position of Rs 96 crore, a long pound sterling position of Rs 36 crore, a short Japanese yen position of Rs 24 crore and a long United Arab Emirates (UAE) dirham position of Rs 12 crore. The five positions exist. They sit on the books at month 12. Somebody bought or sold something and now the bank is holding the result.

A sensitivity measure is a statement about them. The measure says: if the rupee moves by such and such, the aggregated position is worth so many rupees more or less. That statement can be computed, recomputed, argued about, published, forgotten or never made at all, and through every one of those the five positions sit exactly where they are. Deleting every sensitivity measure in the institution would change what is known about the exposure and would not change one rupee of the exposure itself.

The everyday version is a medical one and it lands immediately. A person's blood pressure is a fact about that person. The reading on the machine is a statement about the fact. Skipping the appointment does not lower the pressure; it removes the number. People understand that instantly about health and lose the distinction the moment the subject becomes financial. Measured exposures then get treated as the real ones, and unmeasured exposures quietly stop feeling like exposures at all.

The distinction matters practically in two directions. Going one way, a bank that improves its measurement will often appear to become riskier. Exposures that were always there begin to show up in a number for the first time. Nothing got worse; the reporting got better, and a board that cannot tell those two apart will punish exactly the work it should be rewarding. Going the other way, a measure that is reduced without the position changing has not reduced anything. Change the assumed behavioural life of a deposit book and the reported figure moves; the deposits have not moved. The simulation below does precisely that, and watching a headline fall to nothing while the balance sheet stays put is a useful thing to have seen once.

How is a sensitivity measure different from value at risk?

The two measures answer different questions, and neither is a stricter or larger version of the other. The distinction is worth being precise about. The two figures sit side by side in the same monthly pack at this invented bank, and a reader who lines them up as though they were on one ladder will draw a conclusion neither of them supports.

A sensitivity measure says: if this one variable moves by this much, the change in value is this. The variable was chosen. The size of the move was chosen. There is no probability anywhere in the statement, and there is nothing about any other variable. Rs 4.2 crore a basis point is a complete answer to a question somebody asked.

A distributional measure says something structurally different. The one day trading book value at risk at this invented bank is Rs 15.6 crore, produced by historical simulation on the Rs 3,600 crore held for trading book, and it is a statement about a range of daily outcomes rather than about any move somebody nominated. Nobody chose a variable. Nobody chose a size. Everything that can move is allowed to move together, and the answer describes where a bad day sits within the spread of days. One measure answers a question somebody posed; the other describes a range of outcomes nobody specified, and the two are not points on a single scale of severity.

Two housekeeping notes before either figure is used, and both are the sort of thing that causes real confusion in real meetings. First, they are measured on different objects. The Rs 4.2 crore a basis point is measured on Rs 84,000 crore of rate sensitive positions across the whole balance sheet; the Rs 15.6 crore of one day trading book value at risk is measured on Rs 3,600 crore of held for trading positions and on nothing else. Comparing their sizes without naming their objects is comparing two answers to two questions about two different books. Second, the digits need care: the one day trading book value at risk of Rs 15.6 crore is an entirely different object from this invented bank's cumulative one year repricing gap of Rs 15,600 crore, and the two figures share their digits and share nothing else. Which one is meant has to be said, every time.

How the value at risk figure is produced, what the three methods do differently, and what it means when realised losses run past it are all covered separately in this sequence.

What does a 1.0 per cent currency move do on this same balance sheet?

Exactly the same machinery, a different variable, and one result that lands somewhere most readers do not expect. The currency case is worth working because it shows that a sensitivity measure is a class rather than a single technique, and because the two answers it produces are a lesson in reading the destination as well as the size.

First the aggregation. A currency position cannot be measured until the position itself has been decided. The five positions FX1 to FX5 include both long and short exposures. Add the long side: Rs 144 crore plus Rs 36 crore plus Rs 12 crore gives Rs 192 crore. Add the short side in absolute terms: Rs 96 crore plus Rs 24 crore gives Rs 120 crore. Vindhya Commercial Bank Limited measures its aggregate net open position as the greater of those two, Rs 192 crore, and that Rs 192 crore runs against the Rs 240 crore cap of limit L7, at 80.0 per cent utilisation. Adding the five algebraically instead gives Rs 72 crore, and that figure would report a bank holding five separate open currency positions as very nearly flat. The shorthand method this follows is published by the Bank for International Settlements at bis.org, and the standard that binds in India comes from the Reserve Bank of India at rbi.org.in. The open position as an object in its own right is covered separately in this sequence.

Now the sensitivity. A 1.0 per cent move in the rupee against the aggregated position of Rs 192 crore is Rs 1.92 crore. Same three decisions as before: the variable is the exchange rate, the move is 1.0 per cent, and everything else is held still. The Rs 1.92 crore is a real result and it reaches the profit and loss account when the positions are marked.

Now the same move on a different exposure at the same institution. The bank has a net investment of Rs 480 crore in one overseas branch. A 1.0 per cent move on that branch investment is Rs 4.8 crore, two and a half times larger than the first answer. And it does not reach profit at all. The Rs 4.8 crore moves the translation reserveThe line inside equity that carries the effect of restating an overseas balance already held, which is why a move there does not reach profit.. Restating a balance the bank already holds is a reporting effect on a holding it has not sold, rather than cash changing by a different amount than expected. Two sensitivity figures of similar size, computed the same way on the same day at the same institution, land in two entirely different places in the accounts, and the measure alone does not say which.

The everyday version: a household with money in a foreign bank account and a household with a foreign course fee due next month are both exposed to the same rupee. The first sees the value of something already held restated. The second finds out how many rupees it actually has to hand over. The sizes may be similar. The consequences are not remotely the same, and the difference is worked in full elsewhere in this sequence.

Can the answer be predicted before anything moves?

The simulation below has two controls. Working the default out on paper before either is touched is worth more than any number watched appearing on its own. A prediction committed to in advance tests understanding; a number watched arriving does not.

Two things here reward attention. First, running the shock from 1 to 400 without touching the duration. The three bars keep their shape exactly: the difference stays at 16.7 per cent of the asset side at every single setting. All three quantities are the same balance multiplied by the same shock, and only the durations differ. Scaling a shock cannot change the shape of the answer. Put another way, the per basis point figure already contained everything the 200 basis point figure reports.

Second, set the shock back to 200 and walk the liability duration along. At 2.00 years the answer is Rs 1,680 crore, being 169.7 per cent of the Rs 990 crore cap of limit L8 and a comfortable breach. At about 2.4107 years it is Rs 990 crore exactly, sitting on the cap. At the locked 2.50 years it is Rs 840 crore at 84.8 per cent. At 3.00 years the duration gap closes to zero and the answer is nothing at all. At 3.50 years the gap is minus 0.50 years and the value of equity rises by Rs 840 crore under the same rise in rates, the same size in the opposite direction. Nothing on the balance sheet moved through any of that.

The second table read with the first one beside it carries the central point. The shock control changes how big the answer is. The duration control changes the answer itself, including its sign, and it does so without a single rupee moving anywhere on the balance sheet. A sensitivity measure is only ever as solid as the input it was built on, and on this balance sheet the two large numbers cancel so nearly that the whole reported result lives inside the difference between two estimates. Rs 840 crore is 16.7 per cent of the Rs 5,040 crore the asset side alone moved, and that ratio is the most honest single statement about how much confidence the headline deserves.

What can a measure that moves one variable never see?

Everything that happens between two variables. Not approximately, not with a wide error band, but not at all, because the measure was never asked the question. Vindhya Commercial Bank Limited has a worked case of somebody forgetting that, and it is worth walking through slowly. The interesting part is that no arithmetic went wrong anywhere.

In month 3 of this invented year, on days 21 and 22, the measured one day trading book value at risk at Vindhya Commercial Bank Limited went to Rs 19.2 crore and then Rs 18.6 crore, against the Rs 18.0 crore cap of limit L5. Month 3 carries breach B5, and the locked cause is a widening between two government bond maturities that the bank's model had been treating as offsetting each other. One position had been bought against the other on the understanding that the two moved together, so the pair reported as very nearly flat. Then the two stopped moving together.

Now ask what a basis point value on that book would have said in the days before. A basis point value would have said the position was nearly flat, and it would have been correct. Basis point value moves the general level of rates and reads off the change in value. If one maturity gets richer and the other gets cheaper by the same amount, the general level of rates has not moved at all, so the measure has nothing to report. The variable that actually moved, the gap between the two, was one of the things being held still. The measure was not wrong about breach B5, it was silent about it, and silence and a reading of zero look identical in a report.

The household version of this is a couple who insure the house and the car with the same insurer to get a discount, and then measure their insurance risk by asking what happens if the house burns down and separately what happens if the car is written off. Both answers are correct. Neither of them contains the case where the insurer itself fails, and that is the one case in which the two are not independent at all. Measuring the two things one at a time never produces the number that only exists when they are considered together.

Who actually uses a sensitivity number, and what do they do with it?

A measure with no user is a number in a spreadsheet. Four quite different people at and around this invented bank pick up Rs 4.2 crore a basis point and do four different things with it, and watching what each one does is the fastest way to see why the measure is built the way it is.

The first is the treasurer. Devendra Achar, invented, runs treasury at this bank, and what he wants from the figure is a size. If he judges that the value of equity is more exposed to a rise than the board is comfortable with, he needs to know how much protection to arrange, and the per basis point figure converts directly into that. Rs 4.2 crore a basis point on the whole balance sheet is the yardstick against which any offsetting position gets sized. An offset that changes the figure to Rs 3.0 crore a basis point has done a measurable and stateable amount of work. Sizing is the great practical virtue of the one at a time design: it turns a vague worry into a quantity that a hedge can be measured against. Which instrument he would use is a separate matter: a swap, a forward and a bond are covered under fixed income and derivatives.

The second is the asset and liability committee, G4 at this bank, and that committee sets the behavioural assumptions the measure rests on. The committee does not argue about the arithmetic. The arithmetic is one multiplication and nobody disputes it. The committee argues about the inputs, and in particular about how long the Rs 36,000 crore of current and savings balances behaves as though it lasts. The judgement moves the liability duration and therefore the whole answer. The second control in the simulation above shows exactly that. With the liability duration moved from 2.50 years to 3.00, the headline goes from Rs 840 crore to nothing at all, with no rupee anywhere on the balance sheet changing hands. A committee that understands that is arguing about the right thing.

The third is the committee that sets the caps, G2 at this bank, and that committee turns the measure into a control. A number that nobody can breach is not a control, so somewhere the per basis point figure gets multiplied by a chosen scenario and set against a cap: here that is the bank's own 200 basis point internal move against the Rs 990 crore cap of limit L8, at 84.8 per cent utilisation. Notice the chain of choices that has now accumulated: a duration the bank estimated, a scenario the bank picked, and a cap the bank set. Every one of the three is the bank's own, so a utilisation of 84.8 per cent is a statement about this institution's own policy and never a statement about what any authority requires.

The fourth reader stands outside. An analyst comparing two institutions wants the per basis point figure precisely because it is the only one of the three numbers that is nearly comparable. Caps differ, scenarios differ, and utilisation percentages are therefore not comparable across two institutions at all. A change in value per basis point, divided by the size of the balance sheet, is at least the same question asked twice. Even there the comparison is rough. The durations underneath came from two different sets of assumptions.

And the everyday version. The arithmetic is identical at every scale. A household holding a fixed deposit and carrying a home loan is running the same mismatch as this bank in miniature: an asset whose worth is fixed for a term and a liability whose cost resets. If somebody in that household ever asks what a quarter point move is worth to them in rupees a month, they have computed a sensitivity measure. The household has chosen the variable, chosen the move, and held its spending pattern still, and the answer tells it how much of a change to care about. It does not tell them whether the change is coming.

Where do these standards come from, and what must an Indian bank compute?

Everything so far has been jurisdiction free. A balance, a duration, a move and a multiplication would work identically in any country. None of them is a rule: the mechanism is arithmetic, and arithmetic is not issued by anybody.

The framework a bank's market risk measurement sits inside is published by the Bank for International Settlements at bis.org. The Basel market risk framework and the interest rate risk in the banking book standard come from there, together with the standardised interest rate shocks, the outlier test and the shorthand method for aggregating a currency position. The Bank for International Settlements is the origin. The framework is where the shape of the discipline was agreed, and it is worth reading in the original rather than in summary.

The rule that actually binds a bank in India is a different question with a different answer, and the answer is the Reserve Bank of India at rbi.org.in. Which book a position sits in and which measurement approach may be used are both set there. So are the contents of the return, the recipient, the frequency and the date from which all of it applies. A reader who can recite the origin and not the rule has learned the subject without learning the job, and the two lists are genuinely different documents held by genuinely different bodies. The Indian Banks Association at iba.org.in carries banking operational convention in India.

The size of any shock an authority requires, the level of any outlier threshold, the value of any multiplier, any cap on a behavioural assumption, and the date from which any of it takes effect are all set by the bodies named below, and each must be confirmed there. The 200 basis point parallel move used throughout is Vindhya Commercial Bank Limited's own internal scenario, invented, and it is not a requirement from anywhere. So is the Rs 990 crore cap of limit L8, so is the 3.00 year asset duration, and so is the 2.50 year liability duration. Every one of them is the invented bank's own working number, and every genuine figure of that kind must be confirmed at source before anybody uses it for anything.

What this establishes, and what follows from it

A sensitivity measure can now be built from two inputs, and what it cannot see can be stated. Concretely: the measure is three declared decisions and one multiplication; basis point value is a balance times a modified duration times the size of the move; Rs 84,000 crore at 3.00 years gives Rs 25.2 crore a basis point on the asset side and Rs 84,000 crore at 2.50 years gives Rs 21.0 crore a basis point on the liability side; the Rs 4.2 crore a basis point difference scales to Rs 840 crore at this invented bank's own 200 basis point scenario, being 84.8 per cent of the Rs 990 crore cap of limit L8; the same Rs 840 crore falls out of the duration gap route as well, and that is a check and not a second figure; and a measure built one variable at a time is silent, rather than approximate, about two variables moving apart.

One habit is worth carrying above the others: whenever somebody hands over a sensitivity figure, the question to ask is what was held still. The question is short, it is never rude, and the answer tells more about the number than the number does.