Insurance

Individual pricing based on risk

December 2007

The idea that premiums should reflect the actual risk assumed might sound a little obvious. But it is only in the last 15 to 20 years, with the revolution in processing power that accompanied the introduction of the desktop computer, that even the largest insurance companies have truly been in a position to base pricing fairly and squarely on a realistic assessment of risk.

Until quite recently the cynical view that insurers' pricing is simply a question of guesstimating a figure and adjusting up or down, depending on how the amount paid out in claims compares with the amount received in premium, was not too far from the truth. When I first joined the insurance industry mainframe computers spewed forth reams of paper detailing all claims received for the year. A venerable underwriter went through these manually, sorting claims by postcode area (i.e. the first three or four characters of the postcode). Rates would simply go up in areas where there were more claims than expected and down in those with fewer.

Not only was this approach geographically crude, it was also essentially reactive rather than predictive, relative rather than absolute. A shift from responding to events to a pricing model based on a scientific evaluation of risk is at the heart of what we call risk-based pricing. It would still be fair to say that some insurers are a lot more scientific than others - but the most advanced companies in today's market are at the point where they can accurately reflect all the various 'perils' to which an individual property is exposed - fire, theft, flood, storm, subsidence and so on - in an objectively derived overall premium.

Flood cover

By way of illustration, let's look first at flood cover - a topic much in the news this year. The average desktop PC today packs more processing power than the computers that got man to the moon and back, and flood risk is probably the best example of how advances in computer technology have helped transform major insurers' ability to model hazard exposure. Pulling in data from a wide variety of sources, we have now created highly sophisticated flood-risk modelling tools. Based on detailed digital terrain maps, these enable us to assign a specific level of exposure (low, moderate, high or extreme) not just to full postcodes but to single properties.

This degree of precision can make a big difference if you are the owner of a house at the uphill end of a street with a flood-prone river at the bottom! It is also light years ahead of the approach still employed by some insurers whereby entire postcodes are effectively written off simply because they include properties that have flooded. Employing a more sophisticated analysis avoids unfairly penalising householders on higher ground simply because they share a postcode with other properties built on a flood plain.

Although harder than flooding to anticipate, storm risk too is increasingly well understood by the UK's leading insurance companies. Research data available from companies who specialise in studying such phenomena enables us to model the probable paths of storms across the country - anticipating where storm damage could occur and its likely frequency and scale of magnitude. Meteorological records and predictive modelling confirm that storm activity is not purely random but tends to conform to certain broadly predictable patterns as severe weather tracks across the landscape.

Claims experience

Past experience, and more specifically the pattern of claims reported, continues to exert a major influence on most insurers' pricing. But, unless the insurer in question has a sufficiently large share of the market in a given area to yield a statistically meaningful sample, claims data can leave some applicants unfairly penalised. A single insurer's claims experience should always be set in context with all other relevant information available so as to distinguish emerging trends from temporary aberrations.

A good example of this is theft risk. Historically insurers have tended to overreact to individual high-value household thefts. Yet the fact that an opportunistic burglar with a van may have cleaned out one house on a particular street does not necessarily mean that all its neighbours will suffer a similar fate. As with any other peril, a balanced view must take in a whole range of factors including the insurers' own claims data, local crime statistics, housing density, customers' age, the sums insured, security measures undertaken and so on.

Risk accuracy

The insurance industry today is highly competitive and insurers' profit margins have become increasingly narrow. This makes it more important than ever for insurers to put themselves in a position to price their 'product' both realistically and competitively. A company like Royal & SunAlliance has the necessary scale of operations to apply significant intellectual resources to this task. My own team of around 90 people includes 20 statistical experts and actuaries whose role it is to quantify and analyse patterns of risk across the personal lines insurance spectrum.

Their analysis - combined with the experience and insight of the underwriters they support - enables us to understand to an unprecedented degree of accuracy the precise characteristics of the risks we take on. We can judge this by comparing our predictive models (and the pricing they inform) with the pattern of claims experienced in reality - an exercise we carried out recently in relation to the summer floods in Tewkesbury, Sheffield and Hull, where our models and the actual pattern of flooding closely coincided.

Achieving this degree of accuracy across the personal insurance spectrum depends not just on understanding more generalised factors such as exposure to flooding, storms or locally elevated crime levels - but on gathering as much significant information as possible from prospective policyholders. The key limitation here is the patience of the applicant. Ask too many questions and you risk losing what could have been a model customer. We are also keenly aware that our product has to be easy for intermediaries to sell to their clients.

So the challenge is to gather the maximum amount of significant information with the minimum number of questions. It is all about asking the right questions - questions that actually shed light on the most significant factors affecting each specific type of insurance quoted for. It is equally important that prospective policyholders understand and believe that answering insurers' questions has a purpose and a benefit in securing them the best and fairest premium for an appropriate level of cover.

Peril analysis

The following illustration may help to communicate this concept. Let's assume for argument's sake that the average policyholder will account for £100 in household insurance claims each year. As an insurer, we now have the technology and the analytical expertise to determine the relative contribution of each peril to this total. Flooding might account for £10, theft £20, for example, and - perhaps surprisingly to the uninitiated - escape of water (i.e. damage caused by overflowing washing machines, burst pipes etc.) could amount to £40. More obscure perils like impact from animals or objects falling from planes might account for just fractions of pennies.

So we understand, in the abstract, the average contribution of each of these perils to the total cost of claims. But, of course, there is no Mr Average. Every policyholder is different - and the leading insurance companies now have systems and the expertise in place to understand this. So for the first time we are truly in a position to assign a risk-based price for each component of the risk assumed.

Putting this into practice means that the contribution of each peril to the total premium charged - and by extension that total itself - will vary significantly from one policyholder to the next. For the occupants of a new-build on low-lying land, flooding might well be a major component - much less so for a hill dweller. Fire risk would contribute significantly more to the premium of someone living in a thatched cottage than to that of the owner of a slate-roofed terraced property. Subsidence would contribute more to the premium of an older property built on clay soil than a new house built on granite - and so on.

To get an idea of how we would arrive at a risk-based price for a particular peril, let's take the example of accidental damage. The key considerations here would be policyholder's age, their previous claims history, their postcode, property type, age and method of construction, and the number of bedrooms. But we would also take a range of other factors into account, including time left unoccupied, nature of tenancy, alarm details, cover level, sum insured, excess level and whether or not the property would typically be unoccupied during the day.

Statistics

To help us understand the relative significance of all these factors we are continually drawing in statistics - from our own underwriting and claims data, from publicly available sources and from an array of government agencies and independent consultancies - and crunching the numbers to pick out trends and make sense of evolving patterns of risk within our society. But statistics need handling with caution: the data tells us that most people die in hospital - but shunning hospital is an uncertain strategy for cheating death!

Insurers have been doing their best to make sense of statistics of various kinds for many years, beginning with the life insurance sector where the actuarial profession first arose. But where once the level of analysis in the general insurance market was relatively basic, the industry is growing more and more adept at teasing nuance out of the numbers. To take one fairly recent example: it became clear some time in the early 1990s that women represent a better insurance risk than men - for motor claims in particular. But a subtler pattern has only recently emerged. Middle-aged women are only marginally less prone to accident than middle-aged men - but young women much less so than young men - a fact reflected in the sobering statistic that a major cause of death among young women is crashing in cars driven by young men.

The picture that emerges from the data depends very much on how you slice it and what interpretation you apply. We are always looking to understand how one piece of information correlates with another. The figures may tell us that people living in major cities drive smaller cars. But we also know that city-dwellers are on average ten years younger than the general population. So is the difference in car size a function of city living or a reflection of the fact that younger people tend to drive smaller cars?

Unravelling these correlations - separating causes from effects - is part and parcel of the risk-based pricing process and requires a continual dialogue and debate between statisticians and underwriters. Every new impression that arises from the data must pass the test of common sense and experience before we act on it.

Challenges ahead

Although the UK's leading insurers now have an immeasurably better understanding of risk than ever before, there is always room for refinement. There will undoubtedly be challenges ahead - not least the huge uncertainties around the issue of climate change. Allowing for potentially massive alterations in the UK's weather patterns could prove the ultimate test for our sophisticated new modelling techniques. But here at least insurers with a global network can draw some advantage from experience gained in territories more routinely exposed to extreme weather.

The possibly linked phenomenon of flooding - specifically as it affects the growing proportion of housing stock built on low-lying ground - is another major challenge for the future. The solution here, though, depends as much on our success in lobbying the government to fund additional flood defences as on better understanding the risks involved.

The limited space available here allows only a very superficial view of a complex and continually evolving topic. But hopefully this brief introduction conveys a flavour at least of what risk-based pricing is all about. In reality it is not so much a single discipline as the art of drawing together and interpreting many different strands to achieve an essentially simple - yet previously elusive - goal: the fair and accurate calculation of premiums.

Nathan Williams is underwriting director for personal insurance at Royal & SunAlliance