Modeling approaches to claims reserving in general insurance
Here's how to make accurate forecasts for 'Incurred but Not Reported' (IBNR) claims.
Reserving is an important business function, and its goal is to set aside money for paying out claims.
In the reserving process, claim adjusters determine the ultimate value of a claim. To do so, they deploy learning from the past cases, and the settlement they had for similar cases, and use this information in their estimation of reserves.
The ability to correctly predict the final claim amount is key for insurers and has significant impact on financial statements, as the reserve amount is reported in Quarterly Earnings statements. Technology and computing power has improved exponentially over the years and can help in reaching an accurate reserve number.
However, as it stands today, insurers face challenges of over–reserving and under-reserving. These challenges are what actuaries and risk managers grapple with in their quarterly assessment, which they have to explain to the board.
A good reserving assessment is a sound financial and operational excellence strategy.
The opportunity cost to over-reserving can’t be underestimated. A conservative approach by risk managers and others in the value chain can have a compounding effect. It can significantly increase the loss forecast estimates, and will have an impact on the reinsurance premiums.
Claim reserving challenges
Claim reserving has two major components. The first one is case reserves for Reported but Not Settled (RBNS) claims, and the second one is case reserves for Incurred but Not Reported (IBNR) claims for which the reporting occurs later and is usually more difficult in estimation.
For RBNS, actuaries are challenged in their reserve assessment by the variety of contracts and covers like accident and health, motor vehicle, aircraft, shipping, goods in transit, property damage, general liability, and so on. Some of these claims are complex and take time to settle.
However, claims like property losses due to fire are easily estimated and quickly settled. The IBNR claims such as product liability may be settled long after the policy has expired.
Another category of claims that complicates the process of IBNR provisioning is latent claims, which are reported long after the occurrence of an event. For example, workers may inhale asbestos while performing their jobs but may not file a claim until after being diagnosed with an illness 40 years after the adverse event occurred.
Managing outstanding IBNR claims
To calculate insurer profits, it is very important to calculate the best estimates of the amounts on outstanding IBNR claims that will be paid off later. This can be done either by predicting claims frequency and severity separately or considering the overall impact.
Traditionally, reserving is done via triangulated approaches using actuarial such methods as:
- Chain Ladder: This method operates under the assumption that patterns in claims activities in past will continue to be seen in future.
- Expected Loss Ratio: This method requires a measure of exposure which is usually the earned premium. Loss ratios are investigated and applied to premiums to derive ultimate claims. This a naive method and is used when data is scanty, unreliable or missing.
- Bornheutter-Ferguson: This method is a credibility estimate method based on weighted average of the expected level of claims (estimated by expected loss ratio method) and projection of ultimate claims based on experience to date (estimated by chain ladder method). It is a combination of the above methods mentioned above.
- Projected case estimate: This method relies on case estimates projections and paid to outstanding factors (ratio of payments against case estimates) to derive projected payments.
After analyzing the results from the methods listed above, actuarial judgment is applied to arrive at claim reserving figures. However, the triangulation approaches are less reliable for cases where past trends are not a true reflection of future and in cases where estimating loss ratios from methods such as the expected loss ratio method is difficult. In such conditions, exposure-based methods are used where more analysis is done into expected exposure and industry/market trends and associated factors are used to derive IBNR reserves.
Estimating reserves for IBNR claims
Case reserves are set by claims officials; regular discussions are held with actuaries regarding their adequacy and estimation methodologies. A typical IBNR reserve setting process using triangulated approaches is a four-step process led by actuaries, it involves:
- Data preparation
- Actual vs expected analysis
- Selection of assumptions & methodologies
- Making recommendations & reporting of results
Actual versus expected analysis is carried out to check the credibility of previously selected assumptions/methodologies. This is done to ensure that the projected values are close to the actual values. A large deviation signifies the low credibility of the model, and in these conditions, the assumptions need to be revised.
After conducting actual versus expected analysis, assumptions for building the predictive model are laid out. This includes setting up the inflation rate based on either the price index or the average weekly earnings. Most often this selection of assumptions/methodologies depends upon the availability of data and also the geographical location in which the insurer has sold its product. The projected payments are then adjusted based on the inflation. Discounting is another useful approach which can be used to adjust the projected payments. It involves setting assumptions about the interest rate.
To get the projected payments, we run the data in all models and select the best model based on judgment. The judgment depends upon whether the portfolio has run off or if the insurer is expecting any more claims in the future. An ensemble approach using more than one models can also be used, and weights can be assigned. For example, an insurer can take 25% of future outstanding claims values from chain ladder method and 75% from the BF method. Using such an approach helps in spreading the risk and reducing the error part.
When past trends fail to reflect the future
Exposure-based methods are used in cases where past trends do not reflect a true picture of the future. Exposure-based methods are based on using factors responsible for changes in losses from one year to next and are based on data used and not any judgment or assumptions, and thus present a better approach to forecast their impact on losses.
The basic steps include projecting the number of claims using a probability distribution. The distribution is selected by studying the exposure which projects the future claims. Estimate the proportion of claims which will be non-zero, and then estimate the cost associated with it again using past experience, market statistics, probable maximum loss etc. Ultimately inflation & discounting assumption is applied to arrive at our present value. This is especially applicable for medical reserving practices like asbestos claims discussed at the start of the article.
Finally, based on the output of the selected methodology and assumptions made, actuaries make their recommendations for the provisions that should be booked in the balance sheet.
The case for great data
Currently the actuarial reserving exercise for IBNR claim is generally performed at the aggregate level by the line of business or the coverage type. The best method or combination of methods to be used in particular situation may be depended upon a variety of factors and actuarial judgement. However, LOBs where the factors responsible for change in insurance losses (year on year basis) have significant bearing on future events, a data-driven exposure-based approach is more preferable.
Parin Kalra (Parin.Kalra@exlservice.com) is a Senior Manager of Services at EXL Analytics, a provider of data analytics solutions to financial organizations including P&C Insurance firms.
See also: