Model Governance and Model Risk
The organizational discipline of tracking every model a firm relies on, who owns it, how it was validated, and when it was last reviewed — because a model that's wrong, or simply used outside the conditions it was built for, is itself a source of financial loss.
A trading firm of any size ends up relying on dozens or hundreds of models — pricing models, risk models, signal generation models, portfolio optimizers — each built by a different team, at a different time, under different assumptions. Model risk is the possibility that one of these models is simply wrong, or that it's being used in a situation different from the one it was validated for, and that this error leads to a real financial loss. It's a distinct category of risk from market risk or credit risk: the market can move exactly as a model predicted and the firm can still lose money, if the model itself was flawed from the start.
What governance actually adds
Model governance is the set of processes that catches this before it becomes a loss: a central inventory of every model in use, a named owner responsible for each one, a documented validation process it went through before deployment, and a schedule for periodic re-review, since a model validated correctly five years ago on the data available then may no longer be appropriate for markets that have since changed. Without an inventory, a firm can genuinely lose track of which models are actually load-bearing for trading decisions — a spreadsheet built by an analyst who has since left the firm can quietly become a critical dependency that nobody is actively monitoring or has the context to fix if it breaks.
A concrete case
A firm's options desk relies on a volatility model built and validated years earlier, under market conditions where volatility rarely spiked sharply. A model governance review, conducted as part of a routine periodic re-validation, flags that the model's performance during recent higher-volatility periods hasn't been checked since it was originally built, and that the original developer has since moved to a different team. The review triggers a fresh validation against recent data, which reveals the model's assumptions no longer hold as well in the current environment — a problem caught by the governance process itself, well before the model produced a bad trading decision that would have surfaced the issue the hard way.
What this means in practice
Regulated financial institutions are often required to run some version of this process formally, but the underlying discipline is useful for any firm, regulated or not: know which models you depend on, know who's responsible for each one, and revisit them on a schedule rather than only when something has already gone wrong. The cost is administrative overhead; the benefit is catching a stale or broken model during a routine review rather than during a loss.
Model risk is the possibility that a flawed or misapplied model, not a bad market outcome, causes a loss. Model governance manages that risk with an inventory of models, clear ownership, documented validation, and scheduled re-review — catching problems on a routine schedule rather than after they've already cost money.
Further reading
- Federal Reserve SR 11-7, Guidance on Model Risk Management