A Central Bank Paper Sorts AI Risk Into Four Kinds of Shock

A Bank of Canada paper examines how market-perceived AI exposure is associated with extreme gains and losses in the financial sector across five markets, and it uses Expected Shortfall and skewness decompositions to sort those outcomes into four kinds of scenario: AI-sector, macroeconomic, systemic and financial (S2E1, S2E2).

Two features of the design stand out. It treats extreme gains on the same footing as extreme losses, so in this framing a violent AI-linked rally is a tail event as much as a selloff is. And it works from what markets perceive to be AI exposure rather than from a direct measure of AI activity — the object of study is what is priced, and how violently.

The four categories are the real payoff. If the worry is that AI might destabilise finance, it matters whether the extreme move belongs to the AI industry, to the wider economy, to the financial system, or to the sector doing the trading. A single risk number cannot make that distinction. A decomposition at least forces the question to be asked in four ways rather than one.