Liquidity Stress Testing: Designing the Scenario

A liquidity stress test asks one question: if funding stopped arriving and money left faster than usual, how many mornings could the institution keep paying? Designing the test means choosing who leaves, how fast they leave, and what the buffer can actually be turned into. At Vindhya Commercial Bank Limited, invented, those choices produce 34 days. Every factor inside them is the bank's own.

The answer is a number of days and not a percentage. The count of days is the first surprise in this subject, and it is worth sitting with before anything else arrives. A great deal of liquidity work is reported as a ratio and read as a reassurance. A ratio compares a stock to an assumption. A count of days states when the money runs out under an assumption somebody wrote down. A stress test is only ever as severe as the scenario somebody chose, and choosing that scenario is the act everything else rests on. Everything downstream, every ratio, every horizon, every plan for what to do on the bad Tuesday, inherits whatever was decided in the scenario design meeting.

What does a liquidity stress test actually ask?

A liquidity stress test asks whether the cash is there on the morning it is needed. The question is narrower than it sounds, and it is completely different from whether the institution is worth anything. A solvency question compares what the assets are worth with what is owed, and it can be answered slowly, on paper, at leisure. A liquidity question compares what has to be paid today with what can be turned into cash today, and it is answered against a clock. An institution can answer the solvency question yes and the liquidity question no on the very same morning, and the whole of this subject lives inside that gap.

Here is the household version. The shape is identical and much easier to feel. A salaried person with a paid-up flat worth a great deal of money and eleven thousand rupees in the account is comfortably solvent and can still fail to pay the school fee on the third of the month. The flat is worth what it is worth. The flat cannot be sold before Thursday, and the fee is due on Monday. Nobody in that situation is poor, and everybody in that situation has a liquidity problem. A bank sits in that position every working day, at a scale where the flat is a book of net advances of Rs 57,600 crore and the school fee is a depositor asking for money back before lunch.

A liquidity stress testA computation of what happens to an institution's cash when funding stops arriving and outflows speed up, run on a scenario chosen for severity rather than for likelihood. takes that daily question and asks it under conditions nobody has yet observed. Stress testing as a method, being a scenario chosen for severity rather than likelihood, sits in the risk foundations material along with scenario analysis and the reverse stress test that starts from the failure and works backwards. The harder half belongs here: the design of a liquidity scenario in particular, where the thing being modelled is not a loss but a timing.

How is a scenario designed rather than borrowed?

Every liquidity scenario, at every institution, in every country, is made of exactly three decisions. Who leaves. How fast they leave. And what the buffer can actually be turned into while they are leaving. There is no fourth decision hiding behind those three, and every argument anybody has about whether a stress test was severe enough is an argument about one of them. Two institutions can follow the same published standard, give different answers to the three, and run two entirely different tests that both look orderly on the cover.

Take them one at a time. Who leaves is a statement about behaviour and not about contract. A retail savings balance is repayable on demand and mostly stays where it is. A large wholesale deposit placed by a worried treasury desk is repayable on exactly the same terms and can be gone by lunchtime. How fast they leave decides whether the same total departure is a slow bleed or a single bad Tuesday. The third decision, what the buffer can be turned into, is the one people skip, and it gets skipped because it looks like an accounting matter rather than a design choice. An asset counted at its book value and sold in a bad week at a discount was never worth what the buffer said it was worth.

Are a name-specific stress and a market-wide one the same test?

They are not, and running only one of them is the commonest gap in a liquidity programme. A name-specific stressA scenario in which the institution alone is in trouble while the market around it is working normally, so funding is scarce for it and for nobody else. is a scenario in which the institution alone is in trouble: an adverse news cycle, a rating action, an auction of its own paper that does not clear. The markets are open, the prices are normal, and the only thing that has changed is that nobody wants this name. A market-wide stressA scenario in which funding is scarce for everybody at once, so every action that depends on another institution becomes harder at the same time. is the opposite: everybody is short of cash at once, so the market that was going to absorb the sale of securities is thinner, and every counterparty who was going to lend has the same problem.

The two scenarios attack different parts of the same balance sheet, so one test cannot stand in for the other. Under the first, the buffer sells perfectly well and the funding disappears. Under the second, the funding partly survives because everybody is in the same position, and the buffer sells at a discount that was not in the working paper. Vindhya Commercial Bank Limited numbers its three plan stages P1 to P3, where P2 is the name-specific shape and P3 the market-wide one. The bank's response to either is the contingency funding plan, covered separately.

What is a Liquidity Gap, and what does a maturity ladder show?

Before any scenario can be designed, somebody has to know the shape of the balance sheet in time. A maturity ladderA table of every inflow and outflow slotted into time buckets, so that the timing of cash can be read off the table rather than guessed at. is that shape written down: every expected inflow and every expected outflow slotted into a time bucket, and the timing of money is then read off the table rather than assumed. Vindhya Commercial Bank Limited, invented, uses eight buckets numbered LB1 to LB8, running from one to fourteen days at the front to over five years at the back. Inside each bucket, outflows are subtracted from inflows and the difference is the liquidity gap for that bucket. A gap is a fact about timing and not a fact about quality: a bucket can be deeply negative while every asset in the institution is performing perfectly.

How a balance is slotted is where the judgement enters. Most of the ladder is contractual: a term deposit maturing on day 40 goes into the bucket containing day 40, and a loan instalment due in month 8 goes into the bucket containing month 8. Non-maturity deposits, meaning the current and savings balances that are repayable on demand and mostly are not demanded, cannot be slotted contractually without producing a nonsense in which the entire Rs 36,000 crore of them leaves tomorrow. So they get a behavioural adjustmentSlotting a balance by how it actually behaves rather than by its contract, used because a deposit repayable on demand is not in practice repaid on demand.: the bank places 5.0 per cent of that Rs 36,000 crore, being Rs 1,800 crore, in bucket LB1 and spreads the rest across buckets LB5 to LB8. The behavioural split is set by committee G4, it is the bank's own, and it is the single most load-bearing judgement on the table.

The gap column is where the reading starts. Buckets LB1 to LB5 are every one of them negative, and the negative shortfall in bucket LB1 alone is Rs 2,400 crore, being 25.0 per cent of that bucket's outflows of Rs 9,600 crore. The proportions in the other early buckets are 25.0 per cent for LB2, 20.0 for LB3, 12.5 for LB4 and 16.7 for LB5. Then the sign flips and stays flipped: bucket LB6 is positive by Rs 2,400 crore, LB7 by the same, and LB8 by Rs 4,800 crore. The shape of that column, short money out at the front and long money in at the back, is not an error in the table; it is the description of what a bank does for a living. A bank takes deposits that can leave quickly and makes loans that come back slowly, and it is paid for exactly that.

What does the cumulative gap say, and why is its final zero meaningless?

A shortfall in bucket LB1 may be made good in bucket LB2, or it may pile on top of another shortfall, so a single bucket gap is not much use on its own. The cumulative gapThe running total of the bucket gaps up to a point in time, which is what shows whether a shortfall has been made good or has piled up. is the running total, and at this invented bank it runs minus 2,400, minus 3,600, minus 6,000, minus 7,200 and minus 9,600 across the first five buckets, then climbs back through minus 7,200 and minus 4,800 to exactly zero. Reading the running total as a share of the outflows accumulated so far puts the shortfall in proportion, and that is more useful still: 25.0 per cent through LB1, 25.0 through LB2, then 22.7, 20.0, 19.0, 10.0, 5.9 and finally 0.0.

Now the part that matters. The final cumulative gap is zero by construction and says absolutely nothing about whether the bank is safe. Every rupee of a balance sheet sits in some bucket, so inflows across LB1 to LB8 total Rs 96,000 crore, outflows total the same Rs 96,000 crore, and the last cumulative figure has to be zero whatever the shape in between. A ladder that ties to zero at the end is a ladder that adds up. The information is entirely in the middle: the peak shortfall of Rs 9,600 crore at the one year mark, sitting against cumulative outflows through one year of Rs 50,400 crore, being 19.0 per cent. The peak shortfall, and the speed at which it arrives, is what a scenario has to be designed against.

Why is the same Rs 36,000 crore slotted differently in different tables?

A second treatment of the same money follows, and it would be easy to think one of them is wrong. Neither is. Four tables ask four different questions, so the Rs 36,000 crore of non-maturity deposits, being current accounts of Rs 9,600 crore and savings accounts of Rs 26,400 crore, is slotted four separate ways at this invented bank. The maturity ladder asks when the cash leaves and puts 5.0 per cent of it, Rs 1,800 crore, in bucket LB1. The thirty day coverage computation asks how much leaves in a stressed month and, by applying different assumptions to savings and to current accounts, ends up with a blended 16.2 per cent, being Rs 5,820 crore. The two slottings are 3.23 times apart, a difference of Rs 4,020 crore on one balance, and both sit in the same set of working papers without either being an error.

The other two treatments belong to a different subject and are named here only to place them. The repricing ladder asks when the interest rate changes rather than when the cash leaves, so it places the same balance at bucket RB5, one to three years, and the economic value computation gives it a life of half a year. Interest rate risk in the banking book, the repricing ladder and economic value are covered separately in the market risk material. The discipline that matters here is this: whenever two tables treat one balance differently, the reader has to be told why before the second number arrives, or the second number looks like a mistake.

What is a Liquidity Buffer, and what makes an asset belong in one?

A buffer is a stock of assets held so that a bad month can be paid for out of what is already on the premises. The word doing the work in that sentence is not liquid but available. An asset belongs in a buffer only if it passes three tests on the day, and each test kills a different kind of comfortable assumption. First, can it be turned into cash inside the window, by sale or by borrowing against it, without waiting for a settlement cycle that outlasts the stress? Second, does it hold close to its value while being sold in the same week that everybody else needs cash? Third, is it free: unpledged, unencumbered, not already promised to somebody else, and not required to sit where it is by a rule. A buffer is a claim about availability on the day, not a claim about the quality of an asset.

The everyday version is a household with money in three places: a savings account, a recurring deposit that pays a penalty on early closure, and a security deposit lodged with a landlord. All three are money. Only one of them is available on Monday, one is available at a cost, and the third is not available at all no matter how solid it is. The buffer is the first, plus the second after subtracting the penalty. At Vindhya Commercial Bank Limited, invented, the same reasoning produces four components numbered H1 to H4, adding to Rs 14,400 crore, and the only one that has already been reduced by a discount is H4.

Two things in that table deserve a second look. Components H2 and H3 together are Rs 11,520 crore, being 80.0 per cent of the buffer. The concentration in government securities is not an accident: the paper that holds its value in a bad week is the paper everybody else also wants to hold in a bad week, and sovereign issues are the deepest market most banks have. Government securities, repos and certificates of deposit are defined in the fixed income material; here they are names attached to figures. The second thing is an absence. The balance sheet shows cash and balances with the central bank of Rs 4,800 crore, and component H1 is only Rs 1,920 crore. The missing Rs 2,880 crore is held against the reserve requirement, so it is real, it is cash, and it is not available. The third test was written to catch exactly that.

How to run Treasury Stress Testing: what are the seven steps, in what order?

A liquidity stress test is a procedure before it is a number, and the procedure has seven steps that run in a fixed order. The order is not decoration. A stock valued optimistically and then consumed by a pessimistic flow produces an answer that is wrong in a direction nobody notices, so the buffer has to be valued after its discounts and before the outflow path is run against it. Step one, scope the entity: what legal entity, what currencies, what branches, and whether money can actually move between them on the day. Step two, choose the scenario shape: name-specific, market-wide, or both run separately. Step three, build the outflow path day by day rather than as a monthly total.

Step four, value the buffer after the discounts that would apply in that scenario, not the discounts that apply on a quiet morning. Step five, run the path against the buffer until the buffer is gone, and note the day. Step six, name the actions that would be taken, in what order and by which date. Step six is the point at which a stress test stops being an arithmetic exercise. Step seven, report the horizon rather than a ratio, with the assumptions printed beside it, to the committee that set the appetite. At this invented bank, the assumptions come from committee G4, the appetite clause is A6, and limit L9 is the one that touches the ladder. Committees, appetite clauses and limits, and who accepts a breach of one, belong to the governance material and are handed straight back to it.

Why report a daily path rather than one thirty day total?

Because a total throws away the thing the test was measuring. Under its own severe scenario, this invented bank models net outflows of Rs 720 crore a day on days 1 to 5, Rs 600 crore a day on days 6 to 10, Rs 420 crore a day on days 11 to 20, Rs 240 crore a day on days 21 to 30 and Rs 300 crore a day from day 31 onward. The first thirty days add to Rs 3,600 crore plus Rs 3,000 crore plus Rs 4,200 crore plus Rs 2,400 crore, being Rs 13,200 crore, an average of Rs 440 crore a day. The days the average flattens are the days that decide whether the institution is still open, so the average is real arithmetic and a bad description.

The shape has a reason behind every step of it. The front is fast because the trigger is public and the money that can move fastest moves first: wholesale desks, other institutions, treasurers with a screen in front of them. The middle slows because what is left is stickier, being retail balances belonging to people who have not read the news or have decided to stay. The tail at Rs 240 crore a day is the residue of a stress rather than the stress itself. Then day 31 turns up at Rs 300 crore a day, slightly worse than the tail. The bank assumes that a stress running past a month has started to bring maturing wholesale funding back to the door. Every one of those five rates is the bank's own working assumption, invented, and none of them is a published rate or a requirement.

What does the test produce: a ratio, a number of days, or a decision?

Run the path against the buffer and watch what comes out. The buffer, components H1 to H4, stands at Rs 14,400 crore on the morning of day 1. By the end of day 5 the scenario has taken Rs 3,600 crore. By day 10 it has taken Rs 6,600 crore. Day 12 is the first day on which more than half the buffer has gone, at Rs 7,440 crore against a half of Rs 7,200 crore. By day 20 the running total is Rs 10,800 crore, and by day 30 it is Rs 13,200 crore, or 91.7 per cent of the whole stock. The buffer left on the evening of day 30 is Rs 1,200 crore, being 8.3 per cent, and at Rs 300 crore a day it lasts exactly four more days. The test produces a day: the buffer is gone on day 34.

A day is a different kind of object from a ratio, and the difference is not cosmetic. A ratio compares a stock with an assumption and answers the question is there enough. A count of days answers the question how long, and a person can act on how long. Appetite clause A6 at this invented bank says the bank survives thirty days of its own severe scenario with no recourse to the central bank, so the test does not report a percentage at all: it reports 34 against 30 and the clause holds by four days. The survival horizonThe number of days an institution can meet its modelled outflows out of its buffer before the buffer is gone. as a measure in its own right, including how it is monitored between tests, is covered separately in this material; here it is simply what this scenario produced.

Clause A6 is worth reading for its silences too. The clause does not assume that anybody would step in. Central bank support in a stress, on what terms and against what security, is an entirely separate subject; A6 is a choice the invented bank made about standing on its own, not a statement about what would be available to anybody. The gap and the buffer, both of them above, are different objects: the gap is a structural mismatch read off buckets LB1 to LB8, and the buffer is a stock of assets held against it. The comparison of the two as measures is covered separately, and the scenario above uses both rather than ranking them.

How severe is severe, and who decides that?

Somebody has to choose. The data is a record of days that were not severe, so severity is not discovered in it. Severity is chosen, written down, approved by a committee and then lived with. At this invented bank the assumptions are set by committee G4 and the appetite clause they serve is A6, and what a committee, a charter and an appetite clause are belongs to the governance material. The narrower question is the more useful one: how much would the whole path have to worsen before the answer changed, and if the answer is not much, the scenario was never severe.

The arithmetic is quick. Cumulative consumption to day 30 is Rs 13,200 crore and the buffer is Rs 14,400 crore, so a uniform worsening of the whole path by 9.1 per cent, being 14,400 divided by 13,200, takes the horizon from 34 days to exactly 30 and puts the bank on the edge of clause A6. A worsening of 25 per cent takes it to 23 days. Doubling every day of the path takes it to about 11 days, not to 17. The doubling case is the reason the control below exists: the front of the path is where the stock actually goes, so a fixed stock consumed by a rising flow does not last half as long when the flow doubles.

Move the control and watch the shape of the answer change. Between m of 0.75 and m of 1.25, half a turn of the dial, the horizon falls from 50 days to 23. Between m of 3.00 and m of 4.00, a whole turn further, it falls from 7 days to 5. The relationship between severity and days is a steep curve at the mild end and almost flat at the harsh end. A scenario built by adding a small margin to a comfortable set of assumptions therefore sits exactly where the answer is most sensitive. The steepness is not a reason to choose a harsher scenario for its own sake. The curve is a reason to know, before the meeting, how far the answer moves for how little.

Who actually designs one of these, and what do they do with the answer?

A scenario is not written by a model. A scenario is written by named people in a room, and knowing who they are explains why the design choices come out the way they do. At Vindhya Commercial Bank Limited, four people use the output of the test and every one of them wants something different from it.

The head of treasury designs the path. Devendra Achar at this invented bank is the person who has to say out loud how much of the Rs 36,000 crore of current and savings balances walks out in the first five days, and he is also the person whose funding book the answer judges. The pairing is uncomfortable and every institution has some version of it. The person best placed to know how the money behaves is usually the person the answer reflects on. The assumptions are therefore approved by a committee rather than by the desk that proposes them. The practical defence is not suspicion, it is publication: the five daily rates, the behavioural split and the haircut on component H4 are written down, so a later reader can disagree with a number rather than with a conclusion.

The chief risk officer reads it as a distance to a sentence the board wrote. Sunanda Ravikumar brings 34 days to the table against clause A6's thirty, and the thing she is really carrying is the 9.1 per cent: the whole of that four day cushion disappears if the path worsens by less than a tenth. A risk officer who reports only the 34 has reported a fact. One who reports the 34 and the 9.1 has reported a position. The useful output of a stress test is never the headline number alone, it is the headline number with the distance to failure printed beside it.

An asset liability committee reads it as an instruction about the balance sheet. If the front of the ladder is where the buffer goes, the levers are all at the front: lengthen the maturity of what funds the bank, or raise the stock of what can be turned into cash on the morning it is needed. Which of those a committee reaches for, and who sits on it, belongs to the treasury and governance material.

Someone outside reads it as a claim to interrogate. A lender looking at a corporate borrower such as Nirjhar Industries Limited, invented, is asking the identical question one layer down: how many weeks of committed payments does the cash in hand cover if the receivables slow. A household counting how many months of expenses sit in a savings account is asking it again, with fewer zeroes and no committee. The arithmetic of a survival horizon does not change with the size of the balance sheet: a stock of money, a rate at which it leaves, and the day the second finishes the first. The number of people allowed to argue about the rate is what changes, and a wedding budget with one spender and a bank with eleven committees fail in exactly the same way when the rate is chosen to make the answer comfortable.

Knight, in Risk, Uncertainty and Profit in 1921, drew the line between what can be measured and what merely cannot be known, and a designed scenario sits precisely on it. The 34 days are measured, in the sense that they follow with certainty from two stated inputs. The choice of those inputs is not measured at all. A liquidity stress test converts an unmeasurable question into a measurable one by making an assumption, and the honest version of the output always shows the assumption next to the answer.