A Comfortable Capital Trough Built on a Benign-Calibration Model

Treat 9.8% as an optimistic upper bound, not a central estimate.

Three specific discounts:

1. Calibration. PDs from 2015-2019 embed a benign period with no default cycle. A scenario of -3.4% GDP, 9.5% unemployment and -22% house prices lies far outside that sample, so the model will under-predict defaults, usually non-linearly: PDs tend to multiply rather than rise proportionally.
2. Missing migration. One bucket per quarter for revolving retail under a 22% house price fall is implausible; the modelled loss curve will be too shallow and too late.
3. No overlay against 40% growth in early arrears. That is live evidence the model cannot see, and its absence is a judgement failure, not a modelling choice.

Direction and scale. Higher credit losses and higher RWAs both reduce CET1, so the adjustment is one-directional. Illustratively, on a $1.2bn adverse credit loss base, a 20-25% conservatism uplift is $240-300m - roughly 95-120bp of CET1 on $25bn of RWA. That takes the trough to about 8.6-8.9% before any lag or second-round effects, only marginally above the 8.0% minimum and below the 9.0% board buffer.

Recommend against approving the $300m buyback now.

The arithmetic: 9.8% trough, less roughly 100bp for overlay and model conservatism, gives about 8.8%; less 120bp for the buyback gives about 7.6% - below the 8.0% regulatory minimum and well below the 9.0% board buffer. Even the unadjusted trough leaves only 80bp over the buffer, and the buffer exists precisely to absorb model error. Spending it converts model error into a capital breach.

Conditions under which a buyback could be revisited:

• Overlay and recalibration incorporated, and the re-run trough at or above 9.0% after the buyback.
• A reverse stress test showing the minimum is not breached under a plausible worse case, including a two-notch downgrade and a deposit outflow.
• Committed capital actions covering the next four quarters: scrip or DRIP, AT1 issuance, or an explicit retention plan.
• A board statement that the buyback is discretionary and will be suspended if the trough deteriorates.

Dividend policy should be set from post-stress surplus, not distributable earnings. Approve the ICAAP with a qualified conclusion: adequate only after adjustment.

Failures:

1. Independent validation. A model with no default-cycle calibration, no meaningful migration and no overlay has not been effectively challenged. Validation must sit outside the model owner, report to the board risk committee, and hold authority to withhold approval.
2. Scenario versus model capability. The scenario is severe, but the model cannot express that severity. That gap must be reconciled and overlay-explained in the ICAAP, not left implicit.
3. Use test. An 'adequate' conclusion that is insensitive to the model's own documented weaknesses shows the output is not driving decisions.
4. Overlay governance. No overlay despite a 40% rise in early arrears treats the model as an oracle. Require a documented overlay framework with evidence, quantification and named approvers.

Fixes: commission an independent review within 60 days covering stressed PD calibration, collective provisioning for revolving retail and overlay methodology; re-run the ICAAP with sensitivities; report the revised trough to the board before any distribution decision; and add a board capital metric that shows spot ratio, stressed trough, buffer and distance to minimum side by side, so distributions are decided against the stress rather than the spot number.