Natively non-linear

Second in the series. Last time we covered the founding choice — to absorb complexity rather than simplify it away — and where it started: with structured products. This piece picks up that thread. It's about the first place that choice shows up in the machinery, and the one that matters most: how we handle the many instruments that don't move in a straight line.

Most risk models assume a position moves with the market in a straight line. The market falls 10%, the position falls by some fixed fraction; double the shock, double the loss. For a plain share or a mutual fund, that assumption is close enough to true to build on.

For a large share of what sits in a wealth portfolio, it isn't true — and it fails in the one place that matters most: the rare, severe outcome out in the tail. Modelling that faithfully is what we mean by non-linearity. At Edgelab it isn't a feature bolted to the side of the risk engine. It's the centre of it. That was a deliberate choice, and it's worth explaining why.

It started with structured products

We began, as we said last time, with structured products — among the hardest instruments in finance to value. A structured product is built not to move in a straight line. Its payoff bends: steady income while conditions hold, something abruptly different once they don't. To value one correctly you have to model the bends themselves — the barriers, the conditional coupons, the calls — rather than smoothing them flat.

So the engine was non-linear from the first day. We didn't build a straight-line system and patch in the hard cases; we built for the hard cases. Everything since has been built on top of that.

And this is no niche concern. Structured products are part of the everyday toolkit in wealth management — clients hold them for income, for protection, for a defined payoff. For an advisor, the cost is concrete: you can't responsibly recommend what your system can't measure. So a straight-line engine doesn't just lose accuracy — it narrows the menu of what you can offer, confining you to the simple instruments the engine can see.

Two jobs a straight line can't do

There are two jobs a straight-line approximation simply can't do — and both sit at the heart of risk.

• Tail risk. When a payoff bends, its danger lives in the tail — and the tail is exactly where the straight line is most wrong. Take a reverse convertible: while the underlying holds up, the investor earns an enhanced coupon; if it breaks through a set level, they are left holding the fallen asset instead of their money back. Draw a straight line through that and you sand the barrier reverse convertible away — you delete the single feature that defines the risk. The number looks fine on an ordinary day and is silently wrong on the day you needed it. It's the difference between a dolphin and a shark: from the surface both are just a large fish, and modelling the shark as a dolphin works right up until the day it bites.
• Hedging. A hedge is only as good as your picture of how each side moves. Straight-line a position that's actually curved and you'll mis-size the offset — then find the gap at the worst possible moment, once the market has moved far enough for the curve and the line to part company. Hedging, like tail risk, only works if you've modelled the bend.

What full repricing actually does

The answer to both is full repricing. Rather than stand a product in for itself with a straight line or a handful of sensitivities, we value it as written under each scenario we test. Not an approximation of the product, but the product itself, revalued, with its risk read off the full range of what it would be worth.

We described the overnight run last time: roughly 1.5 billion calculations, finished before sunrise, the answer waiting in about half a second when you ask. What we left out is that the cost of a single pricing varies enormously — a vanilla equity prices in about a tenth of a second; a complex multi-underlying structured product can take minutes. So the work runs on two tracks: a heavy overnight batch for the expensive instruments, and a lighter on-demand path for live questions.

One piece of the puzzle

Full repricing isn't the whole of how we measure risk. The scenarios we run, the universe of factors behind them, and the data underneath all matter just as much, and each has its own article ahead. But it's the piece that decides whether everything above it is built on the real product or on a convenient sketch of one. Get this wrong and the rest inherits the error — which is why it comes so early in the series.

Complexity has a price

Nothing here comes for free, and it's worth being honest about the cost.

Repricing every instrument under every scenario is expensive — less in money than in data. To re-value something honestly, you need its full specification: every clause, every threshold, every date, not just a label. And you need the market data to drive the calculation across every scenario, for the whole universe of instruments. That is a vast amount of information to gather, clean and keep current — which is its own discipline, and its own article later in this series.

The straight-line approach is popular for a reason: it's cheap, and it's fast. We took the harder path for the same reason we always do — a convenient answer and a true answer are not the same thing. The complexity is real. We just keep it on our side of the line.