Risk & Loss Ratio

Risk, as defined by Loss Ratios, is the main metric of insurance data.

A fundamental concept of insurance is Risk, which runs through every aspect of the Insurance workflow. Risk is usually defined by the Loss Ratio. For the sake of business analysis, Risk can be seen to be equivalent to the Loss Ratio, assuming care is taken to acknowledge assumptions, and to balance inevitable shortcomings.

Loss Ratio is a simple ratio of losses over premiums, but there are many complexities. Loss Ratios are only ever fully known when all claims for a book of policies have been closed and loses may only become apparent after decades. A court case’s final settlement may take many years to be judged and appealled.

Predicting a groups of policies final Loss Ratio can be complex, and it can be further complicated by the rigors of oversight and regulation.

Effect of regulation

Many insurance analyses have to balance perfectly with the ledger for practical reasons. In addition to this many also fall under the purview of SEC regulation. The Sarbanes-Oxley Act (SOX) demands that any process used to make financial decisions must be transparent, explainable, and based on clear professional norms.

While predicting future Loss Ratios will never be perfectly accurate, the Actuarial process has been formed to fit in with the regulatory expectations. While this is protective against uncovered exposure, it also tends to stymie novel, and perhaps more creative, approaches to analytics.

Financial Analysis and/or Business Analysis

There is analytics that is clearly used for financial decisions. This will have to follow SOX regulations, and will tend to be conservative, adhering to well established patterns, and, as such, it will be relatively inflexible. Alongside these analytics is space for a parallel track, which is focused on business analysis. These analytics may shed light on the regulated financial analytics but are not designed to replace them. The business analytics can escape from the regulatory controls.

Useful, but fuzzy, Loss Ratios

The calculation of Loss Ratio can often be simplified as long as care is taken. For the analysis of Risk for the sake of making business decisions (Business Analytics), rather than analysis that needs to balance with the ledger (Financial Analytics), Business Analytics can use fuzzy, but useful, Loss Ratios.

For instance:

1. If it can be assumed that there is no distorting change of the underlying policy selection, and no significant change in development, then different categories (groups) of policy may be compared year by year while ignoring any development, such as the Incurred But Not Reported (IBNR) losses. So, for instance, the simple, unadjusted for IBNR, annual Loss Ratios of Properties with & without Landlords can be compared, and the difference can be attributed to the landlord status (all other things being equal).

2. Where premiums are being allocated at the policy level, and so it is impossible to accurately allocate them to specific Class Codes, we can look to see if Loss Ratios remain consistent after we have filtered out the policies where there is uncertainty, or when we share the premium equally among Class Codes. If they do remain consistent, then we have a reason to be confident in any significant result.

The Business analytics must run alongside the regulated Financial Analytics: The regulators will continue to have their established Financial Analytics.

By escaping the embrace of the regulators, Business analytics can squeeze value from data in useful ways.

Risk modelling in an ideal world

In an ideal world, Risk would be accurately modeled, and so it would also be accurately predicted.

Each factor influencing the Risk for every significant business segment would be well enough understood by the Actuaries & Underwriters so that the predicted Loss Ratio outcomes would correlate perfectly with the actual outcomes. The only deviations from the predicted outcome would be those caused by random variations (which would be minimal where there was high frequency & low severity), plus any unpredictable changes in real-world behavior.

So in this imagined perfect world, the actual Loss Ratio of absolutely any grouping, or slicing, of policies should be consistent with a target Loss Ratio (as chosen by the business) whatever the specific analysis. The business appetite for Risk, the Target Ultimate Loss Ratio (TULR), the Predicted Ultimate Loss Ratio (PULR), and the final Actual Ultimate Loss Ratio (AULR) would all match, no matter what the analysis, at any level and with any grouping, varying only through random or unpredictable effects, rather than systemic misalignment of premium with risk.

We can aim to have portfolios that behave in this way. The goal would be to have portfolios whose Loss Ratio would always be consistent with the chosen appetite of the business however the portfolio is sliced or diced. In practice this goal may be useful, even if we know that it may not only be unachievable, but it may also be theoretically impossible.

The Business side would need to set a Target Ultimate Loss Ratios (TULR) for each Review Line (or other aggregation). The TULR will in general be the business wide target Loss ratio, adjusted by the appetite for that specific Line. This is the Appetite made explicit.

The principle is simple, Risk should always equal Appetite, however the Risk is analyzed.

Aggregation at the lowest level

Ideally, the Risk modeling is at the lowest level of the policies aggregation, the Coverage level allowing accurate analyses to be done at any level. Using the lowest level of aggregation also means that as groupings inevitably change the analysis is always valid.

Groupings of policies change all the time. Different departments have different priorities for groupings: Financial may group under Department, Underwriting may group under products, regions or Product leaders while Actuarial want to group policies that share development curves. Ideally all processes, are pushed to the lowest level, the Coverage level, and so there would be no lost information as groupings get broken and remade in ways that the data can not follow.

In the meantime, Business Analysis can lead by analyzing at the lowest level even if the data and can create strategies for data that does not fit. Intelligent forcing will maximize the value of the data, rather than waiting for perfect data. The major advantage is creating a framework of data that is open to flexible analysis.

Problems with Review Line reviewing

Pricing and Reserving reviews and analyses have usually been done at a single level of policy grouping – often called the ‘Review Line’ level. By making these analyses at a single level of aggregation many analytical opportunities are missed. For instance analyzing at a single level can not, in itself, show changes in the book of business.

This lock-in to mid level analysis also means that the inevitable differences in Departmental groupings, or changes in the grouping itself, will cause a loss of useful data as the old data can not follow the new groupings. For instance, if the mix of a Product Line gets adjusted, the old premium history can not be accurately allocated to the new Product.

Underwriters and Actuaries may have their own personal analytics at different levels but these will habitually be siloed, and may be lost when an individual moves responsibilities.

Forcing to the lowest level

In practice modeling on the lowest level of aggregation means that some data, aggregated at a higher level, will need to be forced to the lower level.

The goal should be that analysis is built up from the most granular level – the base coverage level, so that it can aggregate at any level. Some data will already align accurately to the data’s Coverage level such as Losses, and most Class Codes and Risk Addresses but much will not and will have to be forced.

Given that these analyses are designed not to be SOX regulated Financial processes, this can be done with intelligent choices. Forcing data to a lower level does not need to create any significant distortion when appropriate choices are made, and negative effects are monitored.

When there is a concern that results may be negatively distorted by forcing, alternative ways of forcing can be compared. If the alternative forcing methods create results that lead to different conclusions, then the results must be questioned. If all methods align with the same conclusion, then the results become more trustworthy.

Example: Forcing premiums to the Coverage level

Where premiums are allocated to the policy rather than class codes, one may force the data to the coverage lever by a number of methods including:

1) Share all the premium between all the Class Codes

2) Allocate all the premium to the first Class Code. Often the first Class Code is the most significant (Governing Class Code)

3) Filter out all policies with more than one Class Code

If the result of the analysis remains robust to these changes one can be confident the result is valid.

Ideal Risk modeling: PULR = TULR

The significant factors (Risk factors) that influence that Risk should be explicit, and quantified, using historic loss histories and 3rd party data. In this way Loss Ratio analyses (using each significant Risk factor at every level of aggregation) would be visible at every stage of the Insurance workflow.

For every submission, all the Risk factors (internal and 3rd party) plus the desired price for the policy would be used to create a Projected Ultimate Loss Ratio (PULR) which should match the TULR. An intelligent choice of Risk Appetite and maximum insight would be two indivisible attributes of every binding choice.

Automated algorithms can search for, monitor, and triage, significant differences between the TULR (the appetite) and the PULR (the Risk) of any grouping of policies. Any significant difference is a business opportunity.