How effective are the “S4” Series socio-economic tables?

This blog briefly summarises key points from the analysis. CMI Subscribers can find more details in Working Paper 205.

What mortality tables are we investigating? 

In 2024, we published the “S4” Series mortality tables. The S4 tables were based on data for 2014 to 2019 that we received from self-administered defined benefit pension schemes (that is, not those managed by an insurer) with more than 500 pensioners.

 There are 42 different mortality tables, each relating to a subset of the total dataset. For this latest analysis, we are investigating the tables that are defined based on members' socio-economic status:

• Amount band tables were constructed by splitting the dataset based on members’ pension amounts.
• IMD tables were constructed by splitting the dataset based on a combination of members’ pension amount and the Index of Multiple Deprivation (IMD) decile.

For most of the analysis, we projected the relevant mortality tables using the core parameterisation of the CMI_2023 Model.

What data do we analyse?

In this paper, we analyse a large dataset covering 2016 to 2023, comprising over 250 data submissions (with 500 or more members). We split our analysis by pensioner type, considering male pensioners, female pensioners, and female dependants separately.

We only consider data submissions that are in our IMD dataset (a subset of our total dataset that contains data with good quality IMD decile and region data) so that we can also analyse the effectiveness of the IMD tables.

The data we hold is provided by actuarial consultancies, the Pension Protection Fund (PPF) and the Government Actuary’s Department (GAD) and covers both private sector and public sector schemes.

How do we assess the efficacy of the tables?

To assess the efficacy of various S4 tables, we calculate the 100 actual/expected (100 A/E) results using three different measures for the expected deaths:

1. All: Using the all-amounts tables (i.e. S4PMA, S4PFA, or S4DFA) for all members
2. Amount banded tables: Using the appropriate amount band table for a member’s pension amount (e.g. S4PMA_H)
3. IMD tables: Using the appropriate IMD table for a member’s pension amount and IMD decile (e.g. S4PMA_G1).

If the 'Amount banded tables' and 'IMD tables' are effective, assigning members to the relevant tables would lead to results that are closer to 100 (as the tables would better reflect the mortality of the membership, and the expected deaths would be closer to the actual deaths experienced by the pension scheme).

Which tables are the most effective?

For male pensioners, the 'Amount banded tables' and 'IMD tables' measures are better predictors of actual mortality than the 'All' measure. Chart 1 below shows that the median 100 A/Es on these two measures for pension schemes with a reasonable quantity of data are closer to 100 than when assigning all members to S4PMA (111). Similarly, the interquartile ranges of the results are narrower.

The results in Chart 2 for female pensioners are less clear. However, there is still some evidence that assigning members to the relevant socio-economic table improves the predictive ability of the tables.

     
Our analysis highlighted some modelling limitations for female dependants – in particular, there are no female amount banded tables for members with small pensions, so we expect the amount banded tables results to appear heavy. The 'IMD tables' measure appears to provide a slightly better fit to actual experience than assigning all individuals to S4DFA.

What other observations can we draw on the effectiveness of the tables? 

• The efficacy of all three measures and across all three pensioner types improves as the volume of data analysed increases, as expected. Chart 3 shows an example of this for female dependants.
  
  
• We cannot draw any strong conclusions on how the relative efficacy of each measure changes across calendar years, though we note there are materially different volumes of data in each year.
• We have considered the results on an unsmoothed mortality projections basis (described in Section 4.4 of the paper) and come to similar conclusions as for the results using the Core CMI_2023 projections. 

What are the Committee’s future plans?

We will continue to collect pension scheme data and are planning to publish our next annual analysis of mortality experience (covering data from 2017 to 2024) by the end of the year. 

We will also continue to produce one-off analyses, like the investigation into the efficacy of the S4 tables, and encourage CMI subscribers to suggest any topics that they would like to see us investigate.