Recovery indicators and triggers: calibrating so they fire while options exist

What indicators and triggers are#

Recovery indicators are the metrics an institution monitors to detect deterioration in its own condition, and recovery triggers are the thresholds at which crossing an indicator obliges a defined response. The distinction between the two words matters more than it appears: an indicator is something you watch, a trigger is something that happens.

A framework of carefully chosen indicators with no obligation attached to any threshold is a monitoring pack, and it will behave like one on the day it is needed — which is to say it will be read, noted, and filed.

The seven indicator families#

A workable framework draws from seven families, and its coverage is one of the first things a reviewer checks.

• Capital: the ratios themselves and, more usefully, the distance to the levels the institution has committed to maintain.
• Liquidity and funding: survival horizon under stress, concentration of funding sources, the cost and tenor at which new funding is available, and the behaviour of the largest depositors.
• Profitability: sustained losses erode capital more quietly than an event does, and they remove the option of retaining earnings.
• Asset quality: migration and arrears, which lead the loss numbers by some distance.
• Market-based measures where they exist.
• Macroeconomic conditions relevant to the institution’s concentrations.
• Operational and reputational events, which is the family most often left out and the one with the shortest fuse.

The case for market-based indicators#

Market-based indicators deserve a defence, because they are the ones institutions most often argue against including. The objection is that they are noisy, that they reflect sentiment rather than fundamentals, and that a sector-wide move says nothing about this institution.

All of that is true and none of it is a reason to exclude them, because they move before the accounting does: funding costs, spreads, share price relative to peers and rating outlooks reflect what counterparties believe, and counterparty belief is what determines whether funding is available next week.

The sensible treatment is to keep them, define them relative to a peer group so that sector-wide moves do not fire an institution-specific trigger, and pair a breach with an assessment obligation rather than an automatic action.

Fire while the options can still be executed#

The governing principle of calibration is a single sentence: a threshold must fire while the options behind it can still be executed.

Thresholds set close to regulatory minimums fail this test almost by construction, because by the time an institution is near its minimum, the actions that would restore it — raising capital, selling a business, arranging funding — require a market and a set of counterparties that the same conditions have already made unavailable.

The purpose of an indicator is to buy execution time. A threshold that does not buy any has been set for the comfort of the person reading the report rather than the use of the person acting on it.

Working backwards from the slowest option#

That principle converts into a method that works backwards.

1. Take each recovery option and record how long it takes from decision to effect, honestly and in calendar time, including any approval that must be obtained.
2. Take the stress scenarios the institution has already built and read off how fast the relevant metric deteriorates along each path.
3. Then set the threshold far enough above the point of no return that the deterioration rate multiplied by the option’s lead time still leaves room.
4. Do this for the slowest option that matters, because the framework must accommodate the action that takes longest rather than the one that is quickest to describe.

The output is a threshold with a documented derivation, which is also the answer to the only question a reviewer will ask about it.

The failure at the other end#

There is a symmetric failure at the other end, and it is more common than institutions admit. Thresholds set so sensitively that they are breached routinely produce an escalation process that everyone learns to route around: the first few breaches are investigated, the next few are explained, and within a year a breach is a paragraph in a monthly pack with a standing comment attached.

Alarm fatigue does not merely waste effort — it discredits the framework, so that a genuine breach arrives carrying the same weight as the twenty false ones before it. Calibration is therefore a two-sided problem, and an institution with no breaches in years and an institution with breaches every month have the same defect in opposite directions.

Graduated escalation#

Escalation should be graduated rather than binary, because the choice between doing nothing and declaring a recovery situation is a choice most institutions will resolve by doing nothing. Three levels work well.

1. A watch level, breached often enough to be unremarkable, which obliges analysis and a note to the relevant committee.
2. An alert level, which obliges a senior forum to convene within a stated period, assess, and decide whether to prepare specific options — including the preparatory work that makes an option executable later, which is the step most often skipped.
3. A recovery level, which obliges the governing body to convene and decide on activation.

Each level should state its obligation in time-bound language.

A level without a deadline is a suggestion.

What a breach must produce#

What a breach must produce is worth stating precisely, because both extremes are wrong. Automatic execution is wrong: an indicator can breach for reasons that do not warrant the associated action, and a framework that forces the action anyway will be quietly disabled the first time that happens.

Pure discretion is equally wrong, because discretion under deteriorating conditions reliably resolves toward waiting for more information.

The workable middle is a mandatory process with a discretionary outcome: the breach obliges the forum to convene within a defined period, assess against defined questions, and record a decision — and a decision to take no action is entirely legitimate provided it is recorded with its reasoning and a date for reassessment.

The record is the control.

An institution that can show a breach, a meeting, a documented decision not to act and a subsequent reassessment has demonstrated more than one that never breached at all.

Monitoring arrangements#

The monitoring arrangements decide whether any of this operates in practice. Each indicator needs a named owner, a stated frequency, and a data source that does not depend on a manual assembly nobody performs in a crisis.

Frequency should match the speed of the risk rather than the calendar of the committee: liquidity and market indicators that can move within a day should not be observed monthly because that is when the pack is produced.

The framework must also answer what happens between scheduled meetings, since deterioration does not wait for a committee date — which means naming who is authorised to convene an out-of-cycle forum and how quickly, and confirming that the authority does not itself require a scheduled meeting to exercise.

Governing overrides#

Overrides are where indicator frameworks quietly die, and they should be governed rather than prohibited. A breach explained away as a data issue, a seasonal effect or a one-off is sometimes correctly explained away — and each such explanation should be recorded with a named author, so that the pattern becomes visible.

The diagnostic question is not whether overrides occur but whether the same indicator has been overridden repeatedly: an indicator explained away three cycles running is either mis-specified, in which case it should be redefined openly, or it is telling the institution something it does not wish to hear. Both possibilities warrant a decision. Neither is served by an unwritten convention.

Maintaining it against a moving balance sheet#

Finally, the framework needs maintaining against a moving balance sheet. Thresholds derived from a portfolio, a funding structure and an option menu that have all since changed are no longer derived from anything. An annual recalibration tied to the planning cycle is the minimum; a material change in strategy, funding mix or group structure should force one regardless.

The most informative maintenance activity, and the least performed, is back-testing: take any period in which the institution’s condition genuinely deteriorated, however briefly, and ask which indicators moved first, which moved at all, and how much warning the framework would have given. That exercise tells an institution more about its indicator set than any amount of comparison against what other institutions monitor.