P3. Stress Testing: Scenarios That Actually Matter

Risk: From Market Risk to Expected Credit Loss, Part 3. Previously: Expected Shortfall.

Stress testing exists because statistical risk measures can only describe what their data contains. VaR and expected shortfall both read the past and extrapolate. Stress testing asks a different question entirely: what happens if something occurs that the data has never seen?

Three kinds of stress testing

The term covers three distinct exercises with different purposes. Conflating them is the most common mistake in a first stress-testing framework.

Historical scenarios

Replay an actual episode across today’s book: the 2008 crisis, the 2013 taper tantrum, March 2020. The appeal is that nobody can argue the scenario is impossible, since it happened.

The weakness is structural. Markets restructure, and the next crisis rarely rhymes neatly with the last. A portfolio built after 2008 has often been shaped — consciously or not — to survive 2008 specifically.

Hypothetical scenarios

Construct a coherent future state: a 400 basis point parallel rate rise, a 25% currency depreciation, a regional property correction. These cover risks the historical record lacks, which for a market like Vietnam is most of them.

The difficulty is discipline. A scenario must be internally consistent — rates, FX, equity and credit spreads cannot be shocked independently, because they interact. Building that consistency is where most of the real work sits.

Reverse stress testing

Start from the outcome and work backwards. Instead of asking what a scenario costs, ask what scenario would cost enough to breach capital requirements or exhaust liquidity.

This is the most informative of the three and the least comfortable, because it tends to surface concentrations nobody wanted documented. A reverse test frequently reveals that survival depends on one counterparty, one funding source, or one assumption holding.

Designing a scenario that is actually useful

Most stress-testing frameworks fail not on mathematics but on scenario design. Four properties separate a useful scenario from a box-ticking exercise.

1. Severe enough to hurt. If every scenario passes comfortably, the exercise is theatre. A good set includes at least one that produces an uncomfortable number.
2. Internally consistent. Shock the whole picture coherently. A currency collapse accompanied by stable rates and unchanged credit spreads is not a scenario, it is a spreadsheet error.
3. Specific to this balance sheet. Generic scenarios produce generic answers. A bank concentrated in real estate lending needs a property scenario, not a copy of a European supervisory template.
4. Actionable. Each scenario should end in a decision: a limit, a hedge, a contingency plan. A result that changes nothing was not worth computing.

The emerging-market problem

Standard stress-testing methodology assumes a deep historical record. Vietnamese and comparable markets often do not have one, and this is where a framework copied wholesale from a European supervisor falls apart.

Short series. A market with fifteen years of reliable data has lived through perhaps two cycles. That is not enough observations to calibrate severity with confidence.

Structural breaks. The economy of 2010 and the economy of 2026 differ enough that early data may describe a different system rather than an earlier state of this one.

Thin liquidity. Historical price moves understate stress-period moves, because the volume that would have revealed the true clearing price never traded.

Three practical responses. Borrow severity calibration from comparable markets that have experienced the relevant event. Lean more heavily on hypothetical and reverse testing, where the absence of history matters less. And treat liquidity explicitly as a scenario variable rather than assuming positions can be exited at observed prices.

Where this connects to capital

Stress testing is not a parallel exercise to the measures in the previous two posts — it feeds the same decisions. Supervisory stress tests size capital buffers. Internal tests set risk appetite and position limits. Reverse tests inform recovery planning.

The three together form a ladder of severity: expected shortfall describes the tail you can estimate, stress testing describes the tail you can imagine, and reverse stress testing describes the tail you cannot survive.

That ladder transfers directly to credit risk, where the same logic reappears under different names. The next three posts build the credit side.

Frequently asked questions

How does stress testing differ from VaR?

VaR is a statistical estimate from historical data. Stress testing applies specific scenarios, including ones the data has never contained.

What is reverse stress testing?

It starts from an unacceptable outcome and works backwards to find what would cause it. Consequently it surfaces concentrations that forward scenarios tend to miss.

How severe should a stress scenario be?

Severe enough that at least one scenario produces an uncomfortable result. A set where everything passes comfortably has been calibrated too gently to be informative.

How do you stress test with limited historical data?

Borrow severity calibration from comparable markets, weight hypothetical and reverse testing more heavily, and model liquidity explicitly rather than assuming observed prices hold under stress.

Try it yourself

Run a reverse stress test on a simple book, in this order:

1. Take a portfolio and identify its capital or loss tolerance.
2. Pick the two or three risk factors it is most exposed to.
3. Solve for the combination of moves that exactly exhausts that tolerance.
4. Ask whether that combination is plausible, and what would have to happen for it to occur.
5. Write down one action that would reduce the exposure if the answer is uncomfortable.

Step four is where the value sits. A number that produces no action was an exercise in arithmetic rather than risk management.

Next in this series: the three numbers behind every credit model — PD, LGD and EAD.