Risk & Allocation
Risk you can measure and risk you can only imagine
Some risks come with numbers attached. The ones that have historically caused the most damage did not.
What follows is the working version of measurable and unmeasurable risk: the decisions in the order you actually meet them, with the reasoning attached.
Before you start
- Models describe the risks that appear in historical data.
- The most damaging events are typically outside the range the model was built on.
- Robustness matters more than precision when the distribution is unknown.
Two categories that get one word
Some risks can be described with a distribution built from history: how much prices typically move, how often declines of a given size occurred. Others cannot be quantified because the events are rare, novel or structural, and there is no relevant sample. Both get called risk, and only the first appears in the numbers anyone shows you.
The distinction matters because the two require different responses.
Models describe their own data
A risk model estimates from a historical period and is therefore reliable about conditions resembling that period. Events outside it are not predicted as unlikely; they are frequently absent from the model altogether. This is why sophisticated risk systems have repeatedly failed at exactly the moments they existed for.
Precision within a model is not the same as accuracy about the world.
Correlation is the usual surprise
Diversification calculations rest on how assets have moved relative to each other, and those relationships have changed during severe episodes. Assets that appeared independent in normal conditions have frequently fallen together in the worst weeks. A portfolio that looks well diversified by historical correlation can therefore be less diversified in a crisis.
For most people, this argues for holding some genuinely defensive assets rather than relying on statistical spread.
Design for robustness, not precision
When the distribution is uncertain, the useful question is what happens if you are badly wrong rather than what is most likely. Avoiding leverage, avoiding structures that fail suddenly, and keeping near-term money in cash are all robustness measures. They cost some expected return and buy survival of scenarios nobody modelled.
That trade is more attractive the more you doubt the model.
Fragility hides in complexity
Products that behave predictably in normal conditions and unpredictably in extreme ones are common, and the extreme behaviour is often in the documentation. Anything involving borrowing, guarantees or counterparties has a failure mode worth understanding before holding it. Complexity also makes it harder to know what you own at precisely the moment you need to know.
Simplicity is a robustness feature as well as a convenience.
None of this is a substitute for talking to a clinician if something feels wrong.
Living with unquantifiable risk
You cannot plan for specific unknown events, and attempting to leaves you either paralysed or holding expensive protection against the wrong thing. What you can do is keep the portfolio survivable: diversified, unleveraged, with near-term needs covered.
In practice, a margin of safety in your plan matters more than the accuracy of your assumptions. For structures with complex risks, regulated advice locally is a better guide than any general description.
The takeaway
Ask what happens if the model is wrong, not what the model says is likely.
The version you keep doing is the version that works.
Questions readers ask
Should I try to protect against extreme events?
Explicit protection is generally expensive and often mistimed. Structural robustness, meaning no leverage and near-term money in cash, addresses more scenarios at lower cost.
Are risk statistics useless then?
No. They describe ordinary conditions usefully. They should simply not be treated as a complete description of what can happen.
