500m years of solvency: what nature can teach insurers

Your anniversary is coming up, and you have bought a new outfit for the occasion. Two weeks before, you try it on – and the image in the mirror is not flattering. What are your options? You can either accept that you are no longer a size Medium and buy a new outfit, or you can go on a crash diet, live in the gym for the fortnight that remains and try to squeeze yourself in. Two responses to an unexpected outcome: change your beliefs to fit the world, or change the world to fit your beliefs.

In the mid-2000s, neuroscientist Karl Friston proposed a unified model of our brain: the ‘free energy principle’. According to Friston, our brain builds an inner model of its outer environment – a graph of causal dependencies with their probabilistic outcomes – and uses this to forecast what might happen next. At every moment, our brain is comparing prior beliefs against newly acquired data and computing the gap between the two; an Actual versus Expected (AvsE) inner-model forecast versus outer-world realisation. The AvsE is a prediction error, a measure of surprise.

A small AvsE means that the model is well calibrated, whereas a large one signals that reality and model have drifted apart. The brain will try to minimise the error by recalibrating the model – changing its perception, learning, coming up with new explanations for the observed data – or by acting on the world to force it to fit its prior beliefs. These two levers can be combined but there is no third.

Minimising surprise consumes energy; the larger the error, the larger the energy bill. Evolution favoured organisms whose models were both accurate and flexible, leaving more resources for things that matter: finding food, avoiding predators and holding the group together. They were better at separating the signal from the noise. Prediction, in other words, is the most basic function. It is the precondition of staying alive.

“When facing an unexpected outcome, an insurer has the same two moves as the rest of nature: change the model or change the world”

Predictive modelling in nature

These predictive modelling skills are not unique to humans. Every living thing is exposed to a changing environment and subject to the same imperative.

A squirrel buries thousands of seeds across the forest during autumn to stockpile reserves. During winter it digs them out from memory. The map where the seeds are buried and the target size of the stockpile represent the squirrel’s model of the world.

A sunflower tracks the sun from east to west every day. During the night, it swings its head back to the east before dawn in anticipation, based on the timing of its inner model’s sunrise forecast.

An immune system builds a predictive model of the pathogen environment, remembering microbes it has confronted before and archiving this information to produce antibodies next time. A vaccine is simply an extra portion of training data to enrich its inner model.

Even single-celled organisms such as amoebas, exposed to unpleasant conditions at regular intervals, can detect their rhythm, form an expectation and act on the forecast.

A company’s internal model

This brings us to Solvency II regulations, which allow for the creation of an internal model to assess potential probabilistic scenarios that could affect an insurance company’s balance sheet. Viewed through a neuroscience lens, the internal model looks more like a nervous system than a regulatory artefact: the brain that builds expectations of the world, wired to organs that execute specific functions – underwriting and sales, with claims experience providing market feedback. The model perceives, predicts, and tells the rest of the company how to respond.

The model includes different risk types – underwriting, market, credit, operational – and dependencies. Jointly, these provide a distribution of the change in own funds. The company must be able to withstand a significant amount of surprise, arising only once every 200 years, corresponding to the 99.5% quantile of the aggregated distribution function.

When facing an unexpected outcome such as a significant adverse AvsE in claims incurred, an insurer has the same two moves as the rest of nature: change the model or change the world. Reserving and pricing are these two responses in their purest form.

You can change your prior beliefs by lengthening the claims development pattern or raising the a priori expected loss ratio. You can also change the profitability of future policies by revising the rating structure and carrying out active portfolio management. Often, you need to do both to bridge the potential future performance gap based on observations.

Fat reserves and reflexes

Insurance capital is like animals’ metabolic fat reserves. Animals build up these reserves during summer to absorb the shock of winter; if winter is too long or too cold, the fat shortfall could be fatal. Fat needs to be produced and deposited, so every ounce of fat is an ounce not spent on foraging or offspring. That opportunity cost is the cost of capital, and life prices it as we do: more volatile environments require heavier capital buffers. We see this in camels’ humps, which are fat reserves for long desert crossings.

Different products demand different processing capacities and settlement times. Complex contract wordings require a nuanced reading of policy and market conditions, which takes substantial cognitive effort, and the claims that arise can take years, even decades, to settle.

Sometimes, settlement speed matters more than precise loss estimation. Parametric products radically accelerate settlement and are analogous to reflexes. Reflexes are activated in extreme scenarios, such as ‘fight or flight’, when there is no time to think. The signal fast-tracks through the spinal cord, bypassing the brain. The body commits before the consequences can be assessed – a deliberate trade-off.

Reflexes are fast because their world consists of tracking a single signal against a comparison threshold; there is no deliberation. Parametric contracts are the same, paying on a measured index without interpreting what happened. The simplicity is the speed, the basis risk is its price.

Increasing survival odds

Reinsurance is an effective lever for managing capital consumption and making the internal model more robust. Hummingbirds use stop-loss reinsurance to optimise their capital. Making 50 wingbeats per second, they consume an enormous amount of energy and deplete their reserves so quickly that a single cold night without food could be fatal. If a hummingbird’s inner model forecasts a very cold night, it reinsures itself by dropping into torpor, lowering its metabolism to almost zero to shield it from a catastrophic loss. The peak is capped and it survives the night. However, rewarming at dawn costs extra energy as a premium – and it cannot cede all the risk away, as a torpid bird is slow to rouse and can become easy prey.

A lizard seized by a predator sheds its tail to break free. An insurer under capital or liquidity strain does the same, offloading its run-off books through a loss portfolio transfer – a reinsurance or novation of the claims already incurred. Those are usually composed of long-tail lines of business: environmental and general liability, toxic or discontinued lines whose losses are still open and still developing. However, shedding them is not free. The lizard’s tail stores fat reserves that would have carried it through winter; the insurer’s buyer, likewise, demands a risk margin on top of the best estimate. And regrowth is slow and imperfect – the new tail never matches the original, manifesting as lower stability in the lizard and a lesser depth of expertise in the insurer.

Both hummingbird and lizard act on behalf of their inner models, using capital optimisation techniques to increase their survival odds.

Generational memories

To build an inner model, every organism must keep a clear boundary between itself and the world, separating the signals it generates from outside signals. A dog chasing its own tail has failed to perform that task, not recognising the tail as part of its body.

An insurer faces the same problem. It must separate the marks left by its own past decisions from genuine market dynamics. The reason behind poor portfolio profitability may not be explained entirely by a soft market phase or aggressive competitors’ strategy – it may be the anti-selection trace left behind by last year’s ambitious renewals. Pile on additional rate increases to fix it, and you deepen the anti-selection spiral further. You might not be fighting the market but chasing your own tail.

Every model needs to calculate the unobserved risks – events not in data (ENID). Animals make ENID provisions too but more implicitly. A squirrel does not cache for the winter it expects but the winter it fears, burying more than an average season could require. That instinct is survivorship bias in action, the compressed record of every ancestor that overstored and survived the coldest winters.

The squirrel inherits caution via evolutionary selection. Survivors passed on their risk aversion over generations, engraving it into instinct. Deep time did the calibration, and the squirrel never has to think about whether its ENID provision is adequate.

An insurance company has no such inheritance and no such luxury. Its own historical data omits the event that would end it, having only a record of the storms that it survived. What evolution wrote into the squirrel over millions of years, someone inside the insurance company must supply on purpose – and actuaries are best positioned to make that call.

This is what actuarial judgment is for. It is not merely an extrapolation of the historical series but an ability to think outside the box, imagine what the records cannot show, and understand your data’s biases and limitations. The calibration of the internal model must deliberately consider the generational memory.

Preparing for winter

The use test is the cornerstone of the regulatory framework. The Solvency II Directive insists that the internal model be genuinely embedded in how a company is run – used in daily business decisions, not just for regulatory approval. A model not acted upon is a prediction severed from behavioural consequences. Biology is merciless about such an arrangement; the gazelle that sees the lion and does not run is already dead.

Actuaries must be courageous, deliberate and bold. Next time you calibrate a model, remember what you are really doing. The Solvency II internal model is not merely a regulatory framework dressed up in probability theory but a survival principle that life discovered long before there were actuaries, financial markets or the European Insurance and Occupational Pensions Authority. Evolution sharpened species’ prediction skills over millions of years. Actuarial judgment and discipline must always be exercised to balance observed and unobserved eventualities. The prerequisite for survival is a buffer against the winter we have not yet seen.