Institutional Trading

Value at Risk Tells You the Threshold and Ignores the Cliff

The standard regulatory risk measure states a loss level that will not be exceeded most of the time. It says nothing whatsoever about what happens on the days it is exceeded.

↩ Looking BackPart of the 2020 to 2026 retrospective, written in July 2026. The date below marks the 2023 events this piece revisits, not when it was published, so it draws on everything known through mid 2026.
Nathan Xiang·June 21, 2023

The Statement It Makes

Value at risk answers a specific question: what loss will not be exceeded, at a given confidence level, over a given horizon.

A one day 99 percent VaR of 10 million dollars means that on 99 percent of days the loss should be smaller than 10 million. On the remaining 1 percent of days, roughly two or three trading days a year, it will be larger.

How much larger is not part of the answer. That omission is the entire critique.

VaR tells you where the edge is. It is silent about whether the drop beyond it is one metre or a thousand.

Why It Was Adopted

It solved a genuine problem. A large institution holds thousands of positions across asset classes, and senior management needed a single comparable number.

VaR provided one, expressed in currency, aggregatable across desks, and comparable over time. It became embedded in regulation, in capital requirements, and in internal risk limits, and it improved on what preceded it, which was frequently nothing.

How It Is Calculated

MethodApproachWeakness
Historical simulationApply past return periods to today's bookAssumes the future resembles the sample
ParametricAssume a distribution, compute analyticallyNormal assumption understates tails
Monte CarloSimulate many scenariosDepends on assumed distributions

All three share a dependence on the recent past. Historical simulation uses it directly. The other two calibrate their parameters from it. A calm trailing period produces a low VaR regardless of what exposures the book actually contains.

The Failure Modes

Silence about the tail. Two portfolios can have identical VaR while one loses slightly more than the threshold on bad days and the other loses many multiples of it. VaR cannot distinguish them.

Procyclicality. Low volatility produces low VaR, which permits larger positions under a risk limit. When volatility rises, VaR rises, limits bind, and positions must be reduced simultaneously across institutions using the same methodology. The measure encourages building exposure in calm periods and forces selling in stressed ones.

It is not additive in a useful way. Combining VaR across desks requires assumptions about correlation, and those correlations change precisely when the aggregate matters.

It can be gamed. A position with a small chance of a very large loss and no loss otherwise can have a low VaR. Selling far out of the money options is the standard example: the loss falls outside the confidence level, so the measure does not see it.

The Better Complement

Expected shortfall, also called conditional value at risk, answers the question VaR ignores: given that the threshold is breached, what is the average loss.

It captures tail severity, cannot be gamed by pushing risk beyond a threshold, and has better mathematical properties when aggregating across portfolios. International bank capital standards have shifted toward it for market risk for exactly these reasons.

It has its own difficulty: estimating average losses in the tail requires data about the tail, and by definition there is little of it.

What Institutions Do Now

Serious risk functions treat VaR as one input among several. Stress testing applies specific severe scenarios regardless of their modelled probability. Scenario analysis examines defined events. Concentration limits constrain exposures directly rather than through a statistical measure. And liquidity analysis asks how long unwinding would actually take, which no VaR calculation addresses.

The Bottom Line

Value at risk gives a loss threshold at a confidence level and says nothing about losses beyond it, which is where firms fail. It is procyclical, dependent on a calm recent past, and gameable by strategies that place risk outside the measured region. Expected shortfall addresses the tail question directly, and neither substitutes for stress tests, concentration limits, and an honest assessment of liquidity.

Explore Teen Biz News →