Risk Limit Frameworks

A trader who is right most of the time can still sink a firm. The mechanism is almost never a bad forecast, it is a position that grew larger than anyone measured until the one day it went wrong. A limit framework exists to stop that specific failure: it fixes, in advance, how big any single bet is allowed to get, so the decision is made calmly on a normal Tuesday rather than in a panic on the day it matters.

What gets limited

A firm rarely sets just one number. It stacks several kinds of limits that catch different failure modes:

Limit type	What it caps	Catches
Notional limit	Face value of positions	Simple over-sizing
VaR limit	Estimated loss at a confidence level over a horizon	Volatility and correlation risk
Stop-loss limit	Realized loss on a position or book	A thesis that turned out wrong
Concentration limit	Share of book in one name, sector, or factor	A single blow-up eating everything
Leverage limit	Gross or net exposure relative to capital	Silent leverage build-up
Liquidity-adjusted limit	Size relative to how fast it can be exited	Positions too big to unwind quietly

No single limit is sufficient on its own. A notional limit says nothing about volatility; a VaR limit can understate risk in a correlation breakdown; a stop-loss only reacts after the loss has happened. The framework's value comes from the combination.

How limits cascade

Limits are set top-down and consumed bottom-up. The firm sets an overall risk appetite, splits it across desks, and each desk splits its share across traders and strategies. A trader who wants more risk than their allocation must ask the desk head, who must ask upward again if the desk itself is near its own ceiling.

The arithmetic matters: a desk's limit is not the sum of unlimited trader limits. If Desk A is capped at a VaR of $2m, the traders under it must collectively fit inside $2m, which forces the desk head to actively decide who gets how much rather than letting everyone size up independently.

Worked example

A macro desk has a VaR limit of $1.5m at 99% over one day. Three traders currently show individual VaRs of $0.6m, $0.5m, and $0.5m. Summing them naively gives $1.6m, over the limit, but VaR is not additive across imperfectly correlated books. Using the desk's estimated correlation structure, the diversified desk-level VaR comes out to $1.1m, comfortably inside the $1.5m limit.

This is why a risk desk computes limits on the portfolio, not the sum of parts: two traders can each be within their individual budgets while the desk is fine, or a new trade that looks small in isolation can push the diversified total over the line if it happens to correlate badly with what the others already hold. See Marginal Contribution to Risk for how a risk manager decides whether a specific trade is the one to trim.

What breaks a framework

• Stale limits. A VaR limit calibrated to last year's volatility regime can be far too loose once volatility jumps, letting a book that "passed" the limit take on much more real risk than intended.
• Limit arbitrage. A trader who understands the limit metric can construct a position that scores low on the measured risk while carrying real risk the metric misses, for example a tail-heavy options structure that shows a small VaR but a large expected shortfall. This is why many desks pair a VaR limit with a stress-test or expected-shortfall overlay.
• Slow escalation. If breaching a limit merely triggers a form rather than an automatic unwind, the limit becomes advisory. The framework only works if breaches have a fast, pre-agreed consequence.
• Intraday blindness. End-of-day limit checks miss a position that breached and reverted within the day. See What To Watch On The Risk Screen All Day.