The Loss Event: Recording What Actually Went Wrong

A loss event is one failure recorded as one entry, carrying what happened, the process it came out of, its category, its three money columns and its dates. A loss event has more than one date. The date it began, the date it was discovered and the date it was recorded are different dates, and a loss year built on one is a different year from a loss year built on the other.

What is a loss event, and when does something become one?

A household shows the shape of the problem. A tap has been dripping behind a wall for two years. Nobody notices until a patch of damp appears on the ceiling below, a mason is called, and Rs 40,000/- is spent on the repair. When did that loss happen? The money left the house this month. The failure started two years ago. Asked when the damage occurred, the household gives one answer; asked when they paid for it, another; asked when they first knew, a third. Nobody is lying. There are three honest dates attached to one event, and the household never had to choose between them because nobody was keeping a book.

An institution keeps a book, so an institution does have to choose. A loss eventOne operational failure recorded as one entry in the loss record, with its own dates, category, process and money columns. is one operational failure written down as one row in that record. A loss event is not a category, not a trend and not a theme. The entry records a single failure, described well enough that somebody who was not there can tell what happened, with money attached to it and, crucially, with dates attached to it. The moment a failure becomes an entry, somebody has decided which of its dates the row belongs to, and every figure computed from that record afterwards inherits the decision.

Vindhya Commercial Bank Limited, invented for this material, records thirteen operational risk incidents over twelve numbered months, with month 12 as the reporting date. Incident I13 is the entry to open. A trade finance officer and an external party issued 9 letters of credit against forged shipping documents. The scheme ran for fourteen months, ending in month 8, and was discovered when a beneficiary bank claimed. Gross loss Rs 22.4 crore, recovery Rs 7.0 crore, net loss Rs 15.4 crore. The letter of credit itself is named here; what it is and how it pays belongs to a different subject area.

Consider what a row is not. A row is not the sum of several failures that merely felt related. If the same forged document pattern had been presented on other occasions and refused, nothing was lost on those refusals, so they are not part of this row. A row is not a period either. An outage that lasted 4 hours and 20 minutes is one row, and an outage that recurs every month for a year is twelve rows or one row depending on a decision somebody has to make and write down. One failure, one row, and the rule stating what counts as one failure is part of the record rather than a footnote below it.

What has to be on the entry before it is worth keeping?

A loss record with only two columns, a description and an amount, is a list of expensive surprises. Such a list can be totalled and it can be depressing, and it supports very little else. The fields that make it useful are the ones that allow it to be sorted, grouped, and pointed at a year later.

Here is the entry for incident I13 as this invented bank holds it. Read the label column first, then read what sits against it.

Every one of those fields does work. The description lets a reader who was not there tell what happened. The process field is what turns a list into a map. The category is a label from the seven the Basel Committee at the Bank for International Settlements publishes, used here as a labelling scheme and taught separately. The three money columns keep gross, recovery and net apart so nobody has to guess which one a headline figure is on. The risk ownerThe named person accountable for the risk the event came out of, recorded on the entry so the record points at somebody. field means the entry points at a person rather than at the building. And the three date fields matter most. A careless record collapses all three of them into one.

How many dates does one loss event have?

Three, and only one of them is usually written on the report. The date it began is when the failure actually started, and for a long-running event the start is a period rather than a day. The date of discoveryWhen the institution first knew the failure had occurred, which can be years after it began. is when the institution first knew. The date of recordingWhen the entry was written into the loss record, which is the date most loss years are actually built on. is when somebody wrote the entry into the book.

For most failures these three sit close enough together that nobody thinks about them. Incident I3 at this invented bank was a core banking outage of 4 hours and 20 minutes on a working day in month 3. The outage began in month 3, was discovered while it was happening in month 3, and was recorded in month 3. Incident I9 was a vendor-hosted payment gateway down for 9 hours in month 9, and 48,000 transactions failed. Same story. Incident I6 ran for eleven days inside month 6, when the rate applied to 6,200 term deposits was 25 basis points above the approved card. Eleven of the thirteen incidents behave like this. A short failure puts all three date fields in the same month and hides them there. A record built only on short failures never discovers that it has a dating problem.

Two of the thirteen are not like that at all. Incident I13, category 1, ran fourteen months ending in month 8. Incident I5, also category 1, was a branch officer creating 14 fictitious accounts through which Rs 3.6 crore moved, and it ran twenty two months ending in month 5. Gross Rs 3.6 crore, recovery Rs 0.9 crore, net Rs 2.7 crore. For both of these the three dates are far apart, and the moment they are far apart, the choice of which date the row belongs to stops being a formality.

Where do the two long-running events in this record actually sit?

The record runs over twelve numbered months and month 12 is the reporting date. No calendar year is stated and no month is named anywhere in it, so the count runs backwards. Incident I13 ran fourteen months ending in month 8. Fourteen months inclusive back from month 8 lands on months minus 5 through 8. Months minus 5, minus 4, minus 3, minus 2, minus 1 and 0 fall before the window opens. Six of the fourteen months, being 42.9 per cent, sit outside it. Eight fall inside.

Incident I5 is worse. Twenty two months ending in month 5 runs from month minus 16 through month 5. Seventeen of those twenty two months, being 77.3 per cent, sit before month 1. Five sit inside. Between them the two frauds ran for thirty six months, of which twenty three fall outside the reporting window entirely, and both of them land inside it at their full value because both are dated by discovery.

What does the same year cost on each basis?

One step of the arithmetic that follows is an assumption rather than something the record supplies. The record says how long each fraud ran and what it cost in total. The record does not say how the loss accrued month by month. Spreading it evenly is a straight line the reader draws, and it stands as an assumption rather than as a measurement.

Incident I13 cost Rs 15.4 crore net over fourteen months, and straight-lined that is Rs 1.10 crore a month exactly. Six of its months sit before the window, so Rs 6.60 crore of it belongs outside. Incident I5 cost Rs 2.7 crore net over twenty two months, or Rs 0.1227 crore a month. Seventeen of its months sit outside the window, so Rs 2.09 crore of it belongs outside. Together that is Rs 8.69 crore.

Limit L11 is this invented bank's own cap of Rs 60.0 crore on net operational loss over a rolling twelve months, sitting under its own appetite clause A7. The cap is a decision by this bank's own board and not a requirement from any authority.

The same thirteen events, the same rupees and the same failures produce two honest loss years Rs 8.7 crore apart, being 19.8 per cent of the reported year and 14.5 percentage points of limit utilisation. Nothing about what went wrong differs between them. Only the question differs.

Is either basis wrong?

No, and this is the part people get impatient with. A colleague who says the date of the event is the correct basis and the date of recording is a fudge has not thought it through. Ask that colleague what this bank's operational loss was for the year on the event basis. Nobody can know yet. There may be a fraud running right now that will be found in month 26 and will belong, on that basis, to months 3 through 20. The event basis can only ever be completed backwards, and it is never complete.

The recording basis has the opposite property. The recording basis can be closed at month end. Everything known by the reporting date is in, everything not yet known is out, and next month's figure does not reach back and change last month's. A basis that can be closed is usable for a limit measured over a rolling twelve monthsA measurement window that moves forward each month rather than resetting at a year end., and limit L11 is exactly that kind of limit. A limit computable only in hindsight is not a limit anybody can run a business against.

The recording basis answers what this institution found out this year, the event basis answers what this year actually produced, and the fault is never the choice between them but a report that makes the choice silently. The basis belongs beside the number, every time. Stating the basis is the whole discipline, and it costs one line.

What does the record show before the recoveries arrive?

The dates are not the only field with a timing problem. The money columns have one too, and it is easier to see.

Incident I2 happened in month 2. A settlement instruction was sent twice and Rs 42 crore left this invented bank twice. The duplicated payment is the largest gross loss of the year by a distance, ahead of incident I13 at Rs 22.4 crore. Then Rs 41.4 crore came back, and the entry settled at a net loss of Rs 0.6 crore. On the net basis that is joint smallest of the year, tied with incident I8 at Rs 0.6 crore. Notice that both of those sentences named the basis before the position. On gross this entry is the biggest event of the year, and on net it sits at the bottom of the list. A reader given the rank without the basis has been told something close to the opposite of the truth.

Between the money leaving and the money returning there is a gap, and the record does not date the return. The gap is the recovery lagThe gap between the money going out and any of it coming back, during which the same event reads as a much larger loss.. While the lag is running the entry is completely honest and reads Rs 42.0 crore. The same entry read at two different moments gives two different numbers, so a loss report has an as-at date as well as a period, and a report that states only the period has left out half of what a reader needs.

What does a recording threshold hide?

A small shop shows the same thing. The owner writes down every loss above Rs 500/-: the stolen crate, the water damage, the delivery that never arrived. Below Rs 500/- nothing gets written. A book full of Rs 40/- entries is a book nobody reads. The cut-off is sensible. The cut-off is also invisible in the totals, and asked how many things went wrong last year, the honest answer is not a number of events but a number of events above Rs 500/-.

Institutions do exactly the same thing. A recording thresholdA cut-off below which small losses are not entered individually, which keeps a record readable and removes the smallest events from every count taken on it. keeps a loss record readable and quietly removes the smallest events from every count taken on it. The record does not state a threshold at all, and the absence is worth saying out loud rather than assuming. Every incident count in this guide is a count of thirteen recorded events and not a claim that thirteen things went wrong in the year.

Try two cut-offs on this record to feel the size of it. Neither cut-off is this bank's. The record states none. A cut-off at Rs 1.0 crore would drop incidents I2 and I8, both at Rs 0.6 crore net, removing 2 of 13 entries, being 15.4 per cent of the count and Rs 1.2 crore, being 2.7 per cent of the value. A cut-off at Rs 2.0 crore would drop six entries, incidents I2, I8, I11, I10, I7 and I12, removing 46.2 per cent of the count and Rs 6.9 crore, being 15.8 per cent of the value. Almost half the events, a sixth of the money, and the year still reports as a year.

What makes the record usable a year later?

One field on the entry card above deserves a second look. The process linkThe field connecting an event to the business process it came out of, without which a record of losses cannot become a record of weaknesses. is what stops the record being thirteen unrelated stories. Vindhya Commercial Bank Limited runs nine processes numbered PR1 to PR9, and every entry points at one of them. Incident I13 points at PR5 trade finance. Incident I10, where the collateral valuation feed was stale for 11 working days and 340 loans were wrongly marked, points at PR3 collateral management and valuation.

One entry against PR3 is bad luck. Two entries against PR3 in eighteen months is a sentence about the bank rather than about an incident, and that sentence cannot be written unless somebody filled in the field. Without the process link a loss record is a list of expensive surprises, and with it the same list becomes a description of where this institution is weak. The same is true of the category and the risk owner. Category gives the kind of failure it was on a shared vocabulary. The risk owner gives the person who has to answer for it. The process gives the place to go and look.