Exploring how the Solvency II reforms may impact matching adjustment portfolios

Introduction

Proposed reforms to Solvency II are set out in the HM Treasury (“HMT”) Review of Solvency II Consultation and Discussion Paper 2/22 (“DP2/22") from the Prudential Regulatory Authority (“PRA”), both published on 28 April 2022. The focus of the proposed reforms is on changes to the matching adjustment (“MA”) and risk margin (“RM”). We have provided an overview of proposed changes to the MA in Appendix A.

We analysed the implications of these proposed changes for life insurers (in particular annuity providers for whom both the RM and MA are significant components of the balance sheet) as part of our report¹ prepared for the Association of British Insurers ("ABI"), published on 21 July 2022. The analysis and views presented in this report, while building upon the prior analysis, are solely those of WTW and make no use of the ABI member data provided for our commissioned reports.

This article has been prepared for general information purposes only and does not purport to be and is not a substitute for specific professional advice. While the matters identified are believed to be generally correct, WTW does not warrant the accuracy, adequacy or completeness of this information, and expressly disclaims liability for any errors or omissions. As such, any party placing reliance on these materials does so entirely at its own risk.

The purpose of this article is to explore further the potential impacts of the proposed reforms for the MA. To do so, we will consider three illustrative portfolios and explore how the fundamental spread (“FS”) and MA may change under the proposed “Index-Spread Model” as set out in the HMT consultation and PRA discussion paper. We compare the MA and FS for these three illustrative portfolios at 31 December 2020 and 30 June 2022 under the existing Solvency II regime and the proposed Solvency II reforms as set out in the HMT consultation and PRA discussion paper. We also consider the potential impact of using term-dependent indices within the Index-Spread model rather than all-term indices as suggested in the PRA’s DCE.

Values presented in this article, unless otherwise stated, are as at 31 December 2020 (“YE20”) in order to align with the point in time of the PRA’s 2021 quantitative impact study and the analysis presented in DP2/22. However, noting the significant changes to interest rates and spreads since YE20, we have included a section on present day impacts which considers results at 30 June 2022 (“HY22”) to assess the impact of MA changes in the current environment.

The analysis set out in this article only considers the impact on the MA. For the avoidance of doubt, we have not considered how other elements of Solvency II such as Risk Margin, Transitional Measures on Technical Provisions, Solvency Capital Requirements may be impacted by the proposed reforms.

The PRA’s Data Collection Exercise² (“DCE”) set out a potential parametrisation of the Index-Spread Model within paragraphs 3.13 to 3.17 of the DCE instructions. This potential parameterisation forms the basis of our analysis and is described further below.

Key takeaways

Based on the analysis in this article, we highlight the following key takeaways from our assessment of the proposed Index-Spread Model calibration between YE20 and HY22 on three sample MA portfolios:

• The proposals lead to a reduction in MA both at YE20 and HY22, regardless of MA portfolio composition. The reduction in MA is slightly smaller at HY22 although still significant (17 bps for an average MA portfolio at HY22 compared to 26 bps for an average MA portfolio at YE20). This reduction in MA would lead to an increase in the Best Estimate Liabilities of firms who use the MA.
• For more optimised MA portfolios with increased exposure to private assets as typically used when pricing new Bulk Purchase Annuity business, the proposals will lead to larger reductions (25 bps for NB BPA portfolio at HY22 compared to 33 bps for NB BPA portfolio at YE20).
• It is unclear whether the volatility introduced by the Index-Spread Model is intended and we note that such volatility is against the very premise of the original Solvency II MA design. A regime that exhibits such volatility could encourage procyclical investment behaviour and is unlikely to align with the long-term buy and maintain investment philosophy and HMT’s reform objectives.
• The need to adjust the granularity of the Index-Spread Model calibration to provide a more robust credit risk allowance in the MA suggests further challenges with its appropriateness. The Index-Spread Model cannot easily be adapted to suitably allow for durational differences in credit risk and the differing asset characteristics between corporate and private assets.

Illustrative MA portfolios

Our report published on 21 July 2022 contained analysis on the impacts of the Index-Spread Model on differing asset classes based on a simplifying assumption that all asset classes had the same rating, duration and financial/non-financial asset classification.

To further explore the impacts of the Index-Spread Model and how the MA may be impacted by the reforms, we have built upon the analysis previously presented by considering illustrative MA portfolios and calculating the revised FS more precisely for each asset class in the portfolio.

For the purposes of the analysis in this article, we have considered three illustrative portfolios. These portfolios are:

Whilst Chart 3 in the DP2/22 Annex provides the breakdown of the average spread of each asset class into FS and MA, details of the rating, duration and whether the asset class is treated as financial or non-financial are not provided. For the purposes of our illustrative portfolio analysis, we have made assumptions about each of these aspects.⁵

Figure 1 below provides an overview of the asset allocation under each of the three MA portfolios. The key in Figure 1 has been ordered based on percentage allocation.

Each of the three portfolios considered within our illustrative analysis has an average rating of A and a duration of approximately 12 years. We consider this to be broadly representative of UK MA portfolios on average and note that given the Index-Spread Model uses an all-term index (which has a duration of 10.6 years at YE20), the duration of any individual asset contributes only to the idiosyncratic Z component.

For our analysis, we allow for no term dependent spread contribution with a simplification assuming the basis risk for assets with terms shorter and longer than the index broadly averages out. The term differential in spreads is considered further by our analysis of using term structured indices.

Comparison of results

Based on the illustrative MA portfolios described above, we have calculated the MA and FS under both the existing Solvency II MA approach and the proposed approach as described within the HMT consultation and PRA DP2/22. These are presented within Figure 2 below. While we note that this analysis is at YE20, we have presented similar analysis at HY22 in the next section.

Based on the three illustrative MA portfolios, the reduction in MA under the proposed Solvency II reforms compared to the existing Solvency II approach ranges from c. 22 bps for the vanilla MA portfolio to c. 33 bps for the BPA NB MA portfolio with the average MA portfolio having a c. 26 bps reduction in MA. The MA as a percentage of spread for each of MA portfolios reduces by c.17%-20%.

The difference in FS under the proposed Solvency II reforms between the three illustrative portfolios is primarily driven by the impact of the Z component of the Index-Spread Model combined with the impact of slightly different asset allocations and asset class spreads.

Based on the results set out in Figure 2, we see that the proposed Solvency II reforms reduce MA compared to the existing Solvency II approach and lead portfolios to a higher reduction in MA in bps for portfolios which have greater allocations towards illiquid assets and achieve higher spreads. This is a direct consequence of the construction and parametrisation of the Index-Spread Model based on only a single all-term corporate bond index.

The Index-Spread Model also leads to a reduction in MA as a percentage of spread with the most material reduction in percentage of spread being seen for the vanilla MA portfolio which reduces from 66% to 45%. This lower percentage is more closely aligned to our understanding of the matching adjustment in stress as a percentage of spread in Solvency II internal models however we recognise that the regulator may hold different views on appropriate levels of MA as a percentage of spread under stressed conditions versus a long-term best estimate view. The expected treatment of the matching adjustment in stress were the Index-Spread Model to be adopted is still unclear.

Within Figure 3, we have provided a comparison of the MA and FS by asset class for the average MA portfolio under the existing Solvency II approach and proposed Solvency II reforms.

On an asset class basis using the spreads from the average MA portfolio, the proposed Solvency II reforms result in reductions to the MA ranging from 14 bps for covered bonds to 46 bps for ground rents and student accommodation. We note that covered bonds are likely to represent a good measure of market illiquidity given their typically low default risk. The 14 bps reduction in MA for covered bonds highlights that the Index-Spread Model may not fairly reflect risk and could over penalise certain assets.

As part of our 21 July 2022 report, we produced similar analysis⁶ to that shown in Figure 3 but as a simplification we assumed a single term, 'Credit Quality Step ("CQS") and financial or non-financial classification. The analysis presented in Figure 3 above, while being more reflective of the asset classes being considered, produces conclusions that are broadly consistent with those discussed in our report.

For the average MA portfolio, the MA as a percentage of spread for different asset classes as at YE20 ranges from 26% to 70% under the proposed Solvency II reforms compared with the current Solvency II regime where the MA as a percentage of spread ranges from 58% to 90%.

Present day impacts

The analysis presented in the PRA’s DP2/22, our report on the proposed reforms and that shown above are all based on data as at YE20.

Within this section, we explore how the results set out above may have changed since YE20 particularly given the changes in the UK’s economic environment since YE20. The results presented within this section are based on data as at HY22.

The current FS values have been obtained from the PRA’s technical information for Solvency II firms⁷ while the 5-year average and spot spreads used in the Index-Spread Model have been updated using sources consistent with those used as included by the PRA as indicated in the DCE.

Within the table below, we compare the YE20 5-year average and spot spreads to equivalent values at HY22. The X-component impact below assumes that X is held at 35% and the Z-component impact assumes Z is held at 17.5% in line with the DCE. We have assumed that all individual asset spreads move in line with the index.

As shown by Table 1 above, the 5-year average spreads for both financial and non-financial assets across all CQS have decreased. As the X component of the Index-Spread Model is a proportion of the 5-year average, this decrease in the 5-year average reduces the contribution of the X component to the Credit Risk Premium (“CRP”).

At the same time, spot spreads have increased for both financial and non-financial assets across all CQS. As the Z component is calculated as the difference between the prevailing spread on a specific asset and the spot spread on the reference index, we note Table 1 shows no impact for the Z component. For our analysis, we have assumed that all individual asset spreads move in line with the index⁸. This would be the case where the valuation approach follows an index spread with a fixed margin which is a common approach for private asset mark-to-model valuations.

Given the size of the Expected Loss (“EL”) has remained consistent between YE20 and HY22 and remains unchanged from the current Probability of Default allowance, we would expect to see a reduction in the size of the FS under the Index-Spread Model between YE20 and HY22. The magnitude of this reduction is dependent on the specific asset portfolio mix but also the extent to which prevailing spreads on any given asset portfolio may have also changed.

As shown by Figure 4 above, the FS under the proposed Solvency II reforms for each of the portfolios has decreased between YE20 and HY22. This is borne out of the reduction in the 5-year average spread between these two periods. The MA increases because of the increase in spot spreads between the two periods as well as this reduction in FS under the proposed Solvency II reforms.

In comparison, the FS under the existing Solvency II has remained materially consistent between the two periods with the MA increasing to reflect the movement in spot spreads between the two periods.

The increased spot spreads at HY22 and reduction in FS under the proposed Solvency II reforms lead to an increase in the MA as a percentage of spread compared to YE20 (64% to 70% of spread for our three illustrative portfolios at HY22). Similarly, the MA as a percentage of spread also increases under the existing Solvency II approach (75% to 80% of spread for our three illustrative portfolios at HY22).

The MA as a percentage of spread under the proposed reforms has increased by 13% to 19% between YE20 and HY22 compared to an increase of 6% to 9% under the existing approach. This increase in MA as a percentage of is driven by higher spot spreads at HY22 compared to YE20 as well as a lower FS under the Index-Spread Model because of the reduced 5-year average spread. The difference in MA as a percentage of spread between the existing regime and proposed reforms has narrowed between YE20 and HY22.

This analysis highlights that the Index-Spread model is likely to lead to significant volatility in the FS and therefore the MA because of the X component alone (noting that the analysis above excludes any additional Z component impact which may drive further volatility). While we recognise that the PRA intended to implement a new approach which increases the risk sensitivity of the FS, it’s unclear the extent to which the Index-Spread model introduces unintended excessive volatility to the FS which does not correspond to a shift in the underlying long-term view of credit risk.

Whilst the HY22 result from this calibration of the Index-Spread Model is shown to have a slightly smaller reduction in the MA than at YE20 compared to the existing regime (HY22: 17 bps for an average MA portfolio, YE20: 26 bps for an average MA portfolio), this is simply due to the dynamics of current spreads and 5-year average spreads at a specific point in time and therefore is unlikely to give comfort to MA industry participants that moving to the Index-Spread model would not significantly reduce the existing MA.

Term structure of indices

The analysis presented above, using the parametrisation set out in the PRA’s DCE, is not term dependent as the PRA’s proposed calibration is based solely upon an “all-term” index and doesn’t differentiate between short- and long-term holdings. Within this section, we discuss how an alternative application using term-dependent indices may impact the Index-Spread Model.

We have obtained the 5-year average spreads and spot spreads for varying term buckets using an approach consistent with that used by the PRA to obtain the 5-year average spreads and spot spreads within the DCE.

Within the figure below, we have compared the spot spreads and 5-year average spreads under different term buckets against the results for the all-term bucket. This analysis is presented for CQS 2 (A rated) non-financial bonds at YE20 and HY22 for ease of visualisation however we consider that our findings below hold for different CQS and financial and non-financial bonds.

Within Figure 5 and Figure 6 above, we see that both the spot spread and 5-year average spread for term buckets up to 10 years fall below the all-term spot spread and 5-year average spread. We would expect this to be the case given the duration of the CQS 2 all-term index used at YE20 is 10.6 years and 7.8 at HY22.

This use of all-term is likely to be particularly penalising towards assets with a lower term if a greater weighting is given to the X component than the Z-component as the PRA has set out in the DCE.

For example, if we consider a financial asset of CQS 2 in the 1-3 term bucket which has a spread consistent with the spot spread of this bucket:

• At YE20, using the all-term index in the Index-Spread model leads to a CRP of 42 bps while using the 1-3 term index leads to a CRP of 23 bps.
• Similarly, at HY22, using the all-term index in the Index-Spread model leads to a CRP of 37 bps while using the 1-3 term index leads to a CRP of 20 bps.

We note that performing a similar calculation for assets in the 15+ term bucket, where the 5-year average spread and spot spread are greater than those of all-term, leads to only a 4 bps difference at YE20 and 3 bps different at HY22 with the 15+ term index results being more penal.

Despite the above noted limitations, there are advantages to using all-term indices. Firstly, the term bucket indices are comprised of a smaller number of reference assets and therefore there may be volatility in the spreads over time if the composition of the index changes. Similarly, there are some term buckets (e.g. CQS 1 Financial Bonds) which do not have any data at all and therefore simplifications such as interpolation may need to be used which dilutes the benefits of using term-dependent indices. Finally, the use of term-dependent indices adds an additional layer of complexity and so further analysis may need to be performed to weigh up this additional complexity against the additional value provided.

On balance, as the use of all-term indices is overly penal towards assets with shorter terms, there could be merit in considering the use of term-dependent index data by allowing for some form of term-dependent scaling. However, it may be possible that the parameterisation of the Index-Spread model could be amended such that using all-term indices provides a better approximation to term-dependent indices. This would also minimise the additional complexity from using term dependent indices.

Conclusion

Based on the analysis in this article, we see that the proposed reforms to Solvency II reduce the size of the MA regardless of the composition of the MA portfolio and that the size of the reduction is influenced by the asset mix and spreads obtained within the portfolio. This reduction in MA would lead to an increase in Best Estimate Liabilities for firms who use the MA.

At YE20, the reduction in MA under the Index-Spread Model compared to the existing regime ranges from 21 bps to 33 bps whereas at HY22, the reduction in MA ranges from 15 bps to 25 bps. While we note that this is based on a specific parameterisation of the Index-Spread Model as suggested by the PRA in DP2/22, we consider that this could form the basis of the formal proposal.

This reduction in MA may disincentivise rather than promote investment in private assets. We note that the extent of this reduction is likely to depend on aspects such as the balance between additional spread achieved and the cost associated with sourcing these assets. We anticipate that there will be continued discussion amongst stakeholders following HMT’s response to its consultation expected later this autumn.

While the difference between the MA under the existing regime and the proposed reforms has reduced slightly between YE20 and HY22, the capital release benefits of the overall reforms package for the UK life insurance industry is likely to be dampened compared to YE20. In particular, the higher interest rate environment leads to a decrease in the magnitude of the RM under the existing regime and consequently the resulting benefit from a reduction in RM under the proposed reforms.

Whilst the HY22 result from the Index-Spread Model has a slightly smaller reduction in the MA than at YE20 compared to the existing regime, this is purely a function of the dynamics between current spreads and 5-year average spreads.

This reduction in impact between the existing regime and the proposed reforms suggests that the Index-Spread Model may lead to unintended volatility. For example, if spreads remain higher for a prolonged period which may be likely in a long recessionary period as recently precited by the Bank of England, the 5-year averages would gradually increase and may result again in a larger impact at the point of transition to this proposed approach.

It is unclear whether the volatility introduced by the Index-Spread Model is intended and we note that such volatility is against the very premise of the original Solvency II MA design. A regime that exhibits such volatility could encourage procyclical investment behaviour and is unlikely to align with the long-term buy and maintain investment philosophy and HMT’s reform objectives.

Ultimately implementation of the Index-Spread model has significant challenges when considering a suitable matching adjustment calculation approach which may give rise to unintended consequences and strange outcomes in some economic periods. The limitations of using an all-term index versus the additional complexity of trying to adapt the model to allow for term differentials is a good example of this which is difficult to overcome.

Appendix A - Overview of proposed reforms to fundamental spread

The PRA discussion paper sets out a revised approach to calculating the FS. As defined by the PRA, the revised FS is the sum of the expected loss and the credit risk premium. The PRA defines the CRP as “the uncertainty around the expected loss for which a willing arm’s length third party would demand a premium for taking on the risk.” while the EL is defined as equal to the current Probability of Default under the existing Solvency II MA approach.

The HMT consultation details how the CRP should be calculated using the “Index-Spread Model” as follows:

There are four key components to this Index-Spread Model:

1. The X parameter sets an initial level for the CRP to consider uncertainty around future expected default losses. This initial level does not depend upon the individual assets a firm holds but will be dependent upon the level of granularity of the reference indices used.
2. The Z parameter adjusts the CRP based on whether the current market spread of each asset held is higher or lower than the current market spread of the reference indices. This component seeks to address concerns highlighted by HMT and PRA that the current FS approach incentivises investment in assets with high spreads for a given rating.
3. The n parameter represents the averaging period for the reference index which is used in the X component.
4. X and Z terms both refer to a reference index which would provide the average and prevailing credit spreads needed for the calculation. The PRA envisages that the indices used for the two parts would be the same and indicates that a corporate bond index may be a suitable choice given that this underlies the existing FS calibration and has a sufficiently long data set.

While no specific parameterisation is set out in the HMT consultation or the PRA discussion paper, the PRA’s DCE uses the following parameterisation:

• X = 35%
• Z = 17.5%
• N = 5
• Average spreads, YE20 spot spreads and durations are explicitly defined for Financial and Non-Financial assets by CQS which have been obtained from the “all-term” iBoxx indices at YE20.
• Explicit floors and caps on the X component of the Index-Spread Model are defined.