How models are overseen

Tiering by the cost of error

A model is any method that applies statistical, mathematical or algorithmic technique to input data to produce an estimate or a ranking the group may act on. A spreadsheet that does that is a model. The tier follows the consequence of the output, so a simple calculation that sets a reported valuation ranks above an elaborate system that produces internal management information.

Tier 1 outputs determine an order, a valuation reported to a client or an investor, a capital or liquidity figure reported to the Board, or a screening outcome for a client. The twelve Tier 1 models include the four signal families in Quantitative Strategies, the unlisted valuation approaches used by Private Equity, the group risk aggregation engine and the liquidity forecast.

An owner proposes the tier and the Chief Risk Officer sets it. Where a use changes so that an output would carry further, the tier is raised before the new use begins, not after. Using a model outside its validated scope makes it a new model, and a new model cannot be used until the extension has been validated.

• Tier 1: the output determines an order, a reported valuation, a capital figure or a screening outcome.
• Tier 2: the output informs a human decision without determining it.
• Tier 3: the output is internal management information with no external or position effect.

Validation by people with nothing at stake

The validation unit sits inside the Risk function, reports to the Chief Risk Officer and has no reporting line to any model developer or any desk that uses a model. Six people work in it, three in Zurich and three in Singapore. Their work is not charged to the desks whose models they examine, because a validator paid by the business is a consultant rather than a check.

Validation tests the soundness of the approach, the quality and origin of the inputs, the implementation against the documented design, the behaviour of the model at the limits of its inputs and the accuracy of its outputs against realised outcomes. Tier 1 models are validated before first use and every year after. Tier 2 models are validated every two years. Tier 3 models are self-assessed annually and validated every three years.

Findings are graded. The 2025 cycle raised 44 findings across the inventory. Nine were open at 30 June 2026, all grade two or grade three, and none older than 90 days. Internal Audit reviews the work of the validation unit on a three-year cycle, so that the validators are themselves validated by someone else.

Monitoring and the stop rule

Every Tier 1 model is monitored against thresholds fixed at validation, covering prediction error, input data quality, output stability and, for trading signals, realised return against expectation. The owner and the validation unit review the results monthly. The Risk & Valuation Committee reviews them quarterly. Tier 2 models are monitored quarterly. A breach of a threshold is a finding and follows the finding timetable.

Any Tier 1 model can be stopped within fifteen minutes by its owner, by the Chief Risk Officer or by the head of the desk that uses it, with no further approval required. Each has a documented fallback, either an earlier validated version or a manual process, and the fallback is tested at every revalidation. A stopped model restarts only on the written approval of the Chief Risk Officer.

The output is not the decision

A model output is information. It becomes a decision when a person decides to act on it, and the group has placed that person in every chain. For Tier 1 trading signals, a named person authorises the transmission of the orders the signal proposes. For unlisted positions, the Risk & Valuation Committee adopts the figure the approach produces.

For screening, a compliance officer reads every match the model surfaces and records the conclusion. For deal screening, the ranking orders the queue and does not shorten it. This is why the group can say without qualification that it takes no automated decision about any client, counterparty or employee.

What is reported and to whom

The Chief Risk Officer reports to the Risk & Valuation Committee each quarter on the inventory, the validation calendar, open findings, threshold breaches and any model stopped in the period. The Board receives an annual model risk report and approves the list of Tier 1 models itself, because the Tier 1 list is a statement of where the group has accepted the cost of automated error.

Client reports name the figures and decisions that a model assisted, in the same fixed form used for learning systems. The assistance statement applies to a fifteen-year-old valuation method as readily as to a system trained last quarter, since a client asking whether a machine was involved is asking about the decision rather than about the technique.