Why stress test when you already have VaR?

Everything so far has been statistical: fit a distribution to history, read off a percentile. Stress testing abandons that approach entirely — and that abandonment is the point.

A stress test asks a question with no probability attached: "if this specific thing happened, what would we lose?"

The two questions are genuinely different

Note the last row carefully. Stress testing does not solve the model-risk problem; it relocates it. VaR is limited by the data you have. A stress test is limited by the imagination of the person designing it. Neither limitation is removable, which is why serious risk functions run both.

Why the missing probability is a feature

Dropping probability sounds like a weakness. It is what makes the exercise useful.

Every statistical estimate of a rare event is unreliable, because rare events are — by definition — thinly represented in the sample. Arguing about whether a scenario is a 1-in-200-year or a 1-in-500-year event is arguing about a number nobody can estimate well. Stress testing sidesteps the argument: forget how likely it is; here is what it would cost.

That reframing changes the conversation a risk committee can have. "Our 99% VaR is $8 million" invites the response "fine." "A repeat of autumn 2008 costs us $140 million and breaches our covenants in week three" invites a real discussion about whether the institution could survive it, and what would have to be true beforehand for it to survive.

The two families

Historical replay — take a real crisis and apply its actual market moves to today's portfolio. Concrete, defensible, impossible to dismiss as fantasy, because it happened. Its limit: the next crisis will not be a copy of the last one.

Hypothetical scenarios — construct a coherent set of shocks that has never occurred. Frees you from history's specific sequence. Its limit: it is only as good as its internal consistency and the designer's judgement.

The next two lessons take one each.

Reverse stress testing

There is a third mode, and supervisors increasingly require it. Instead of starting with a scenario and computing the loss, start with the loss and work backwards: what combination of market moves would exhaust our capital, breach our covenants, or force us to liquidate?

This inverts the usual failure of imagination. You are no longer asking "what might happen?" — a question whose answers are biased toward what you already worry about. You are asking "what would have to happen to break us?" and then examining whether those conditions are as far-fetched as you assumed. Institutions are routinely surprised by how ordinary the answer looks.

The honest position

Stress tests do not tell you what will happen. They tell you what your current portfolio would experience under a specific named set of moves. That is a conditional statement, and conditional statements are the strongest thing risk management can offer. Anyone offering more than a conditional statement is describing a certainty that does not exist.

Try it now

1. Pick a simple portfolio you can describe in three or four lines (for example, index ETFs and a bond ETF, with weights). Write down its holdings — you'll shock it in the next two lessons.
2. Below is the full history of a broad index. Find the steepest cliff on it, then Measure it properly: drag across the worst four weeks you can find and read both the percentage and the bar count. That drawdown is your first candidate scenario, and it is not a hypothetical anybody invented.

3. Do the reverse exercise: write the sentence "the loss that would force this portfolio to change is ___" and note what market moves would produce it. Framing, not advice — the threshold is defined by whoever owns the portfolio, never by a course.