How to build a capital plan that survives stress testing

What the plan is for#

A capital plan is a forward projection of a bank’s capital resources and capital requirements across a multi-year horizon, showing the resulting ratios in a baseline case and under adverse scenarios, together with the actions management would take if those ratios approached the thresholds the board has set.

It is the part of an ICAAP that converts an assessment of today’s risks into a statement about the institution’s ability to carry out its strategy.

A capital plan that only ever shows comfortable ratios has not been tested; it has been drawn.

How far forward, and whose plan#

The horizon should match the horizon of the decisions the plan is meant to inform, which in practice means at least three years and usually the full length of the approved business plan.

Shorter horizons systematically flatter the position, because the capital consequences of growth arrive before the earnings do: new lending consumes capital at origination and repays it over the life of the exposure, so a plan truncated at two years shows the cost of expansion without its benefit, or — more commonly — is calibrated in a way that shows neither.

The baseline must be the plan the business is actually running. This sounds obvious and is routinely violated: a finance team builds a capital projection from a conservative internal forecast while the commercial teams work to a growth budget nobody reconciled against it, and the ICAAP then presents a capital position for a bank that is not the one taking the risk.

The test is simple. If the growth rate, product mix, margin assumption and cost trajectory in the capital plan cannot be traced line by line to the board-approved plan, the projection describes a hypothetical institution and its conclusions transfer to nothing.

Building a world that holds together#

Scenario design starts with a narrative, not a parameter set. A usable adverse scenario tells a story specific to this bank — which sector turns, which counterparties are affected, what happens to collateral values, how funding cost responds, what the currency does, what management sees first — and then derives the parameters from that story.

Scenarios assembled the other way round, by selecting plausible-looking shocks for each variable independently, tend to produce internally inconsistent worlds: a severe recession with stable property prices, or a currency collapse with unchanged import-dependent borrower default rates. Reviewers notice, because the inconsistency usually runs in the bank’s favour.

Where the exercise is quietly weakened#

Severity is the judgement most often calibrated backwards. The honest sequence is to define severity from the bank’s vulnerabilities and from historically observed stress, then run it and accept the result.

The dishonest sequence — common enough to be the first thing an experienced reviewer probes — is to establish the severity at which the capital position remains acceptable and describe that as severe.

The diagnostic question is straightforward: what would this scenario have to look like for the bank to breach its buffers? If nobody on the team can answer, severity was never tested. If the answer is implausible, the framework may be genuinely resilient; if the answer is a mild extension of the severe case already run, the calibration is the problem.

Losses are only half of it#

Translating a scenario into a capital trajectory requires both sides of the ratio to move. The numerator falls through credit losses, provisioning, margin compression, fee income decline, valuation losses and any restriction on the earnings that would otherwise be retained.

The denominator rises, because exposures migrate to worse risk grades under stress and standardised or model-based risk weights increase accordingly — a portfolio that consumed a given amount of capital in the baseline consumes materially more in the adverse case without a single new loan being written.

Holding risk-weighted assets flat through a severe downturn is the single most common technical shortcut in capital planning, and it removes a large part of the stress the exercise exists to measure.

When the capital hit lands#

Expected-loss accounting changes the shape of the trajectory as well as its depth. Where provisions are recognised on a forward-looking basis, a deteriorating outlook brings future losses into current earnings before those losses are realised, and exposures migrating between staging categories can produce a step change in provisioning that is large, early and only partially reversed later.

The practical consequence for a capital plan is that the worst quarter is usually not the quarter with the worst macroeconomic reading; it is the quarter in which the outlook turns. Plans that model losses as a smooth accumulation across the stress horizon miss the point at which the buffer is actually tested.

What could actually be done#

Management actions are where credibility is won or lost, because they are the only part of the plan that is a promise rather than a projection. Each action needs four things stated: the capital benefit, the execution time, the conditions under which it would be triggered, and an honest assessment of whether it would be available in the scenario that made it necessary.

Issuing capital instruments into a market that has just repriced the bank’s risk is not a management action; it is a hope. Selling a portfolio at book value during the downturn that impaired it is not a capital benefit; it is a crystallised loss.

Restricting distributions is genuinely available and immediate, which is why it belongs early in the ladder rather than as a last resort. The strongest plans show the stressed trajectory twice — before and after management actions — and explain the difference action by action.

Leaving room to act#

The plan needs a trigger ladder, not a single limit. A structure that works sets a regulatory minimum, a board-set management buffer above it, and one or two intermediate levels at which specific, pre-agreed responses begin — reduced distributions, slower balance-sheet growth, a de-risking programme, a capital raise brought forward.

The purpose of the ladder is to convert a deteriorating ratio into a decision while options are still cheap. Triggers set immediately above the regulatory minimum fail this test, because by the time one fires, the only remaining actions are the expensive ones the market can see coming.

Starting at the end#

Reverse stress testing completes the exercise by asking the question the scenarios do not: what combination of events would exhaust the capital position or render the business model unviable?

Because it starts from the failure point and works backwards, it is not constrained by the imagination that produced the scenario set, and it frequently surfaces a dependency nobody had capitalised — a single funding relationship, a concentration that only becomes visible when two sectors are treated as one, a legal exposure with no natural cap.

It is also the part of a capital plan most often omitted, and the omission is easy for a reviewer to spot.

What the plan changed#

Finally, a capital plan is judged by whether it did anything. The evidence that matters is not the document but the record around it: the board discussion in which a growth target was moderated because the stressed trajectory did not support it, the pricing change that followed a concentration finding, the acquisition deferred, the distribution reduced, the limit tightened.

Reviewers ask for that record explicitly, and its absence is read as the answer to the question of whether the plan is a management tool or a deliverable. A plan that has never constrained anything has not been used, however well it is written.