The Metrics We Use

With this article, we’ll go back to basics and talk about the various metrics and measures to evaluate a service contract book. All measures have value but many must be interpreted in the context of a service contract book.

Policy Year vs. Calendar Year

Data is typically organized by policy year, or the year that the service contract began. This means that results by policy year will reflect the profitability of the operation by the year that policy was initiated, rather than when the claim was paid. This is a good tool for evaluating your underwriting. It is common in other insurance lines as well, typically worker’s compensation.

Another way of looking at the business is calendar year – which is simply the financial results of the year – and the premium earned in that year versus the losses incurred. While this is critical for the financial statement, it is less common when evaluating experience. One reason is that it is computationally complex. In order to evaluate the earned premium for the calendar year, you must earn the premium as of this year and then earn the premium as of one year ago. For example, the table below shows how to calculate the earned premium for calendar year 2012:

Acme Auto Administrators: Policy Year Experience

If you have a data warehouse, and you want to evaluate your book by calendar year, you need to store earnings at multiple valuation dates. For example, each contract should show incremental earnings at the end of each year until it is fully earned.

In general, policy year results will tell you about the differences in the types of risks and the level of premium for each year. Calendar year results may be better at detecting trends. For example, a large increase in claims from Toyotas would be more apparent in calendar year results than policy year.

Most insurance lines are evaluated by loss year, but since time between the date of the claim and the payment of the claim is so small for vehicle service contracts, this measure will not produce any meaningful differences from the calendar year method.

Loss Ratios

If we are examining calendar year data, there is just one loss ratio – the calendar year loss ratio. This loss ratio is frozen – once the calendar year loss ratio is calculated for 2012 it will never change.

However, when we look at policy year data the loss ratio is subject to development. For policy year, there are typically 3 types of loss ratios:

1. Current Loss Ratio = Losses Paid (or Incurred)/Earned Premium to Date

This may be the most common metric to evaluate a book. However, it is subject to two sources of distortion. First, the earned premium is an estimate – it may be too high or too low. Second, since it is based on current premium, it is weighted more heavily to contracts which earn more quickly such as used cars or shorter terms. If those segments are performing better, it may give you some false assurance about your overall book.

2. Future Loss Ratio = Future Expected Losses/Unearned Premium

This is the expected loss ratio of the remaining book. In addition to the two sources of distortion mentioned above, it also is dependent on the accuracy of the future loss estimate. Also, the future loss ratio is just the expected loss ratio on the unexpired portion of the book. It does not contain any losses or exposure for new contracts that will be written in the future.

3. Total Loss Ratio = [Losses Paid (or Incurred) + Future Expected Losses]/Net Written Premium

One advantage of the total loss ratio is that it won’t be distorted by earnings patterns, though it still requires an estimate for the future losses. In our opinion, the total loss ratio is usually the best measure for evaluating a book because it eliminates differences due to shorter earning and longer earning contracts.

Another question is how you handle cancellations. You can ignore cancellations if they are a small part of the book – future cancellations will offset future losses. However, if a significant portion of your book cancels then you want to model this separately. The percentage of refunds to gross premiums for business that cancels may be significantly different than the percentage of losses to gross premiums for active business.

Severity

Severity, or the average loss per claim, is one of the few measures we can calculate which is not subject to earnings estimation. Measuring the severity over time (for example, for each calendar year) can allow you to measure the impact of inflation and the performance of your claims department.

Severity = Losses Paid (or Incurred)/Number of Claims

Whether you use paid or incurred losses (that is include any case reserves for claims) should make minimal difference.

An important consideration is counting claims. Do you count claims which have no payments? Typically, there are quite a few of these claims, which are either denied for coverage reasons or have losses which don’t reach the deductible. We would usually advise excluding these claims since these “closed without payment” claims may fluctuate up or down over time due to changes in claims processing. For example, you will never get a complete picture of claims which settle below the deductible – your system may capture a portion of these which get reported. Therefore, excluding all of these zero pay claims will usually result in a truer picture of the actual trend.

Distributional changes in your book can cause large changes in severities which have nothing to do with inflation or claims settlement. For example, if you have largely written contracts with $50 deductibles and then begin to write $200 deductible contracts, you will see a large spike in your severity – as the smaller claims you previously covered disappear.

Other major distributional impacts are the type of car and type of coverage. If there have been major shifts in any of these, you may want to “normalize” severity. For example, you can calculate severity by deductible level. To look at the overall severity, apply a constant distribution to the severities by deductible for each year.

Let’s explain with an example:

In this example, the average severity falls for a simple average. However, a major deductible shift has occurred to lower deductibles. When the distribution of deductibles is held constant, the severity is rising.

Frequency

Frequency is subject to distortion by the estimation of earned contracts. Earned contracts should be calculated using the same methodology as earned premium. If the underlying data is by contract, you can just add up the percentage earned for each contract.

Frequency = Number of Claims/Earned Contracts

Since earnings patterns are configured so that the losses emerge evenly, there may be periods in the contract where the frequency is higher and the severity is lower (and vice versa). Monitoring frequency can be a great tool for planning your administration needs.

Pure Premium

Pure Premium or Average Cost per Contract

Pure Premium = Frequency x Severity

Pure Premium = Losses Paid (or Incurred)/Earned Contracts

Pure premiums are, of course, subject to the accuracy of your earnings patterns. This is a great measure for evaluating your experience without the impact of any rate action.

Almost all of the measures we have discussed have some limitations. Because it is more difficult to compute, there has been a lack of analysis on a calendar year level. However, calendar year results can provide valuable information in monitoring your experience. It is also, of course, valuable to monitor measures over time to measure the change or trend. Sometimes the direction is more important than the actual measure.

What are some of your favorite measures and metrics to monitor service contract business? Drop a comment below and add to the conversation.