The Second Layer

Why per-trade risk management breaks down at the portfolio level.

Ask most people how they manage risk and you get one number: the percentage of risk they allocate to each position. One percent, half a percent, two percent if they are feeling brave. That number is real and it matters for sure, but it is also only half the job.

I run risk at two levels, and encourage you to do the same. There is a set of controls at the level of each individual strategy, and a separate set of controls sitting on top of my portfolio. They do different jobs. The single biggest jump in how survivable my book felt came when I stopped treating risk as one dial and started treating it as two layers that do not always necessarily have to agree with each other.

What a single layer of risk actually protects

If you size every trade at one percent of equity, you have answered exactly one question: how much does it hurt when this specific trade hits its stop? That is a useful question. It keeps any individual loss survivable and it stops a single bad call from putting a dent in your year.

Here is what it does not answer: what happens when five of your strategies are all long risk at the same time because the signals happened to line up? What happens when a Sunday gap blows through several stops at once? What happens in the kind of week where correlations across FX, indices and metals surge toward one and the diversification you thought you had is all of a sudden just missing?

Per-trade sizing treats every position as if it lives alone. In a real systematic book, positions do not live alone. They share exposure, they share regimes, and on the worst days they share a direction. The single-layer trader finds this out the hard way, usually during their first proper correlated drawdown, when a book full of supposedly independent bets all moves against them together.

The second layer is the one that exists for the day the first layer was never designed to handle.

Layer one: controls at the sub-strategy level

The first layer I build is at the level of each individual strategy. Every sub-strategy in my portfolio has its own risk budget and its own limits, designed and tested before it ever touches live capital.

The key control here is a cap on how much historical drawdown the sub-strategy is allowed to show. I build and size each strategy so that its maximum historical drawdown, in testing, sits around one percent at the sub-strategy level. That is deliberately tight. It means no single strategy in the book is capable of doing serious damage on its own, even in the worst stretch of its own backtest.

Why so tight? Because I am going to run a lot of these strategies together. If I let each one have a four or five percent drawdown profile, then a handful of them aligning in a bad week could put the whole book into a hole I cannot easily climb out of. By keeping each one small and well-behaved, I buy myself the freedom to combine many of them without the combination becoming a monster.

Always remember, a -50% drawdown needs a +100% return to recover. These numbers are not symmetrical.

The sub-strategy layer is also where I do the unglamorous work of strategy hygiene: checking that the live behaviour stays within tolerance of the tested behaviour, watching for the early signs of decay, retiring strategies that drift. A strategy that stops matching its own risk profile gets pulled. That decision happens at this layer, in isolation, before it becomes a portfolio problem.

Layer two: controls at the portfolio level

The second layer sits above everything. It does not care which strategy generated which position. It looks at the whole book as one object and asks a simple question: how much can this entire machine lose right now, across every open position, if conditions turn bad?

This is the layer most systematic books are missing, and it is the layer that matters more than anything. Because usually, the failure mode that actually ends people is not one strategy going wrong. It is many strategies going wrong together, in a way that no individual strategy's risk budget anticipated, because each one was only ever looking at itself.

The portfolio overlay does a few things. It watches total exposure across the book, so that even if every strategy independently wants to be long, the aggregate position is capped at something I can survive. It accounts for the fact that correlations are not static, that the diversification benefit I usually relish shrinks in extreme market environments. And it gives me a hard ceiling on whole-book drawdown that is independent of what any single strategy thinks it is doing.

Why the two layers have to disagree

Here is the part that took me a while to internalise. The two layers are supposed to conflict sometimes. That conflict is the entire value of this design.

There will be days when every sub-strategy is behaving perfectly, every one inside its own risk budget, every one doing exactly what it was built to do, and the portfolio layer still says no. Too much aggregate exposure. Too much correlation. The combined book is offside even though no individual part of it is. On those days the portfolio layer overrules the sub-strategies, and that override is the moment the whole structure earns its keep.

If your two layers never disagree, you have not really built two layers.

If your two layers never disagree, you have not really built two layers. You have built one layer and a copy of it. The sub-strategy controls and the portfolio controls have to be looking at genuinely different things, asking genuinely different questions, so that one can catch what the other structurally cannot see.

Dynamic, not fixed

Neither of these layers is a static number I set once and forget. My risking is dynamic. The amount of risk the book carries adjusts to conditions rather than sitting at a fixed percentage through calm and chaos alike.

The logic is straightforward even if the implementation is not. A fixed risk percentage is too much risk in a volatile, correlated regime and too little in a calm, dispersed one. Sizing that responds to the environment keeps the actual risk being taken closer to constant, even as the market underneath it changes character. The dial moves so that the outcome does not have to.

I am not going to lay out the specific mechanics, partly because they are the bit I would rather keep as a black box, and partly because the specific implementation matters far less than the principle. The principle is that risk in a real book is a moving target, and treating it as a fixed constant is how people get surprised.

How to start building the second layer

If you run a book with only per-trade sizing today, here is roughly how I would add the second layer without over-complicating it.

Start by measuring your real aggregate exposure. At any given moment, add up what your whole book is actually exposed to, net and gross, across correlated instruments. Most people have never looked at this number. Looking at it is uncomfortable and educational.

Then stress your correlations. Do not assume that what you have tested in regards to correlation will stick forever. It will not. Recompute how your strategies behave together in the worst historical windows you have, and size for that, not for the average.

Set a whole-book drawdown ceiling that is independent of your per-strategy budgets. Decide the number that, if the entire machine hit it, would make you stop and reassess. Build a control that enforces it regardless of what individual strategies are signalling.

Make your sizing responsive to volatility. Even a crude regime filter that halves risk in high-volatility conditions is better than a fixed percentage that ignores the weather entirely.

And test the two layers against each other before you trust them. Run your historical data and find the days where the portfolio layer should have overruled the sub-strategies. If it never does, your second layer is not actually independent, and you have more work to do.

The point of all of it

Risk management is not the exciting part of trading. Nobody builds a strategy because they are excited about drawdown ceilings. But the difference between a book that survives a decade and a book that has one great year and then gives it all back is almost always the risk architecture.

One layer protects you from the trade in front of you. The second layer protects you from the version of the market that your strategies, each looking only at themselves, were never built to see coming. I run both because I have watched what happens to people who only run one. I also endured this through April 2026, the 17th April to be exact, when multiple of my strategies went long on Dollar-Yen and my Gold strategies started firing simultaneously. It was a bad day, P&L wise, but more so process-wise. It highlighted vulnerabilities to me, vulnerabilities that I have since acted upon and recommend you do the same before you learn the hard lesson.

Let me answer some questions you may have

Isn't per-trade sizing enough if my strategies are uncorrelated?

They are uncorrelated until they are not. Correlation between strategies is regime-dependent, and it tends to climb in exactly the extreme conditions where you need diversification most. The portfolio layer exists to handle the days your uncorrelated book temporarily becomes a correlated one.

What's a sensible whole-book drawdown ceiling?

There is no universal number. It depends on your strategies, your capital, and your tolerance. The useful exercise is not copying someone else's figure but deciding the level at which you would genuinely stop and reassess, then enforcing it mechanically so the decision is already made before the bad week arrives.

Should sub-strategy limits really be as tight as one percent?

That works for me because I run many strategies together and want each one to be individually harmless. If you run a small number of strategies, your per-strategy budget can be larger. The tightness of the sub-strategy layer is a function of how many of them you are combining.

Does dynamic sizing not just reduce my returns?

It changes the shape of them. You give up some upside in the calm periods in exchange for taking far less damage in the violent ones. For a book you intend to run for years, that trade is almost always worth making, because the violent periods are what end careers.

How do I know my two layers are genuinely independent?

Run your history and look for the days the portfolio layer overruled the sub-strategies. If those days exist, your layers are doing different jobs. If they never happen, your second layer is just a restatement of the first, and it will not save you when you need it.

Thanks for reading.

Kieran