After attending last month’s Insights Session on Australian Disability Income Table, Graham Kelly provides a comprehensive overview of the session that covered implementation, key learnings, strengths and cautions.
The ADI 2007-11 disability income table was released in May 2015. It was the first new industry wide table targeted at Australian disability income business for over 20 years and the first ever to be wholly derived from Australian experience. The life industry has since been implementing the new table. It is providing new understanding of individual disability income business claims experience, profitability and reserving. A very timely and well attended Insights Session was held on 31 August, sharing learnings from practitioners about implementing the table and strengths and cautions when using the table.
The session was divided into three parts:
- Briallen Cummings (KPMG), who led the development of the table, gave an introduction;
- Emily Galer (AMP) and Mari-Lize Hill (Comminsure) led the detailed implementation in their companies, and talked about key practical learnings from the implementation journey;
- Wallace Lee (Munich Re), made some observations about strengths and cautions when using the table. Munich Re developed its own disability income table for Australian business, prior to the availability of the ADI 2007-11 table.
Introduction – Briallen Cummings
Briallen talked first about key principles underlying the table’s development. A generalized linear model was used to provide insights as part of the development, a first for a standard table in the Australian life industry. A very significant judgment overlay was applied to the findings. Everything in the table had to be supportable by general reasoning based on likely influences of claims experience. Nothing was included “just because it was in the data even though we can’t think of why”. This judgment based approach has subsequently been vindicated from data received after release of the table. Another key principle was that the table must be “easy to use, understand and adjust”. This is very important and may seem obvious. However, there is significant judgment in making the trade-off between (a) more granular representation of the experience versus (b) the additional complexity arising from each additional feature introduced into the table. Briallen indicated the judgments made rely to some extent on the mix of business within the industry. For blocks of business with very different mix from the industry, for example all white collar or all blue collar, companies need to be aware the table may not be as effective as a base for representing the experience.
Briallen also gave the results of a survey of industry progress in implementing the table. Fourteen companies responded, of which ten had completed implementation, two were in progress and two had not implemented. The average elapsed time to implement was 5 months, and the average FTE requirement 13 months. Averages are potentially misleading, as there were very large variations between companies. Most of the factors in the table were adopted during implementations, with companies’ own experience generally being allowed for as overlays to the table, rather than more granular adjustment.
Implementation and key learnings – Emily Galer and Mari-Lize Hill
The experiences of Emily and Mari-Lize in implementing the table had many more commonalities than differences. Emily talked initially about the importance of scoping and allowing for all aspects of implementation “end-to-end”. Key “end-to end” components are:
- Amending projection models and data inputs.
- Amending the experience analysis software, process and data.
- Performing the detailed fitting, based on the company’s own experience.
- Analyzing and understanding results.
- Stake-holder management.
She stressed that these components are neither independent nor fully sequential. Learnings and constraints in one will affect the others. Mari-Lize’s experience was similar.
The effort required in amendment of projection models and data inputs is significant. The high level structure of the table, based on an initial scan, is not too complex. However, this is deceptive because implementing in detailed models requires understanding the table to very high degree of detail. For example the synchronization of timing of application of the factors with how the model treats cash flow timing – such as middle or end of month – can make quite a difference to final results. Data may not be available to implement exactly as intended by the table, resulting in a compromise which must be agreed by relevant stakeholders. Time must be allowed for both testing and understanding the results. The latter is very likely to be iterative – changes to the model will likely be required as a result of the process of seeking to understand and analyze the results. The first iteration is unlikely to be the last.
The changes to experience analysis software, process and data must have appropriate consistency with changes to the models and their input data. If this is not in place, there will be an unnecessary and false discontinuity between the past (actual experience) and the future (projected experience), with inaccurate results from the model even if the model itself is “fully correct”. One particular challenge is that, whereas the projection models typically only require a data set at a single current point in time, experience analysis requires data going back for a number of years. Further compromises or approximations may be required.
Fitting is the process of deriving the adjustments to the table so that the final assumptions are appropriate for the company, having regard to its experience. This is perhaps the more interesting or “fun” task. One issue discussed was credibility – disability income business has more product choices and rating factors than other life products, resulting in many areas where individual companies do not have credible data. For this, the approach of “follow the industry shape” was most common. Group business was also mentioned – the table is based solely on individual business data and is designed for individual business. Nevertheless group business often “piggy backs” off the individual business information infrastructure. It was noted that in some cases the table had been applied in relation to terminations for group business, without policy duration as a rating factor and with some calibration to relevant observed experience.
Because implementation of the table can have a significant impact on reported financial results and on views of profitability, stakeholder management to senior executive levels and the broader business is important. One key challenge requiring management is to resist the temptation to report results too early. The implementation of the table is new and more complex than most assumption changes, so more time must be allowed than normal in the project timetable for testing, analysis and understand the results . As is often the case with something new and complex, this process often leads to discovery of issues which must be fixed.
Strengths and cautions – Wallace Lee
The final speaker, Wallace Lee, spoke initially of how most of the shapes and features of the ADI 2007-11 table are consistent with Munich Re’s table. This, effectively independent, validation is not normally available for a standard table and should be seen as a strength of the table.
Wallace then talked about four “cautions”, when using the ADI table:
- It is not necessarily applicable to all blocks of individual disability income business. Wallace felt it did apply to most individual disability income business, but had seen some portfolios where not all the table’s rating factors applied.
- Treatment of ancillary benefits in company data: the table has been constructed on the basis that all payments classified as “ancillary” have not been included in the primary benefit analysis. Each company has its own way of processing and recording these benefits in its data. Wallace stated that the Munich Re table adopted a different approach, whereby all payments in respect of time periods after the end of the waiting period were incorporated in the primary analysis. This had led to relatively higher incidence, but also higher early claim duration terminations, than in the ADI table. There is no “correct” answer to this, with main conclusion being that companies need to consider this in applying the table.
- Classification of benefits as accident or sickness: Wallace noted that this classification is a key foundation of the table, but there is no industry standard on it. He had noted that there is significant variation in the proportions between companies, mostly driven from how they are classifying claims with muscular-skeletal causes. This provides a further reason for companies not to adopt the table shape without validation against company’s own data.
- Variation in the effect of experience drivers over the course of a claim: Wallace talked first about how criteria can be either convergent (decrease in effect as claim duration increases), divergent (increase in effect as the claim duration increases) or flat (effect is relatively uniform over the course of claim). Examples of these within the table are waiting period, occupation and cause of claim (all convergent), benefit period (divergent) and age shape, benefit type and smoker status (all flat). Wallace expressed the view, borne out from his analyses at Munich Re, that no criteria are truly flat. This view, particularly related to age shape, means that the table is less “accurate” than if it used a method which recognizes how age shape varies over course of a claim. He acknowledged this increases complexity, so is part of the trade-off described in Briallen’s introduction.
There were quite a number of questions of the speakers, especially for the practical implementation component. This indicated the session was useful and appreciated by those attending.
CPD: Actuaries Institute Members can claim two CPD points for every hour of reading articles on Actuaries Digital.