3.1. Origins

The Pensions Act 2004 replaced an earlier regime where employers were required to meet a minimum funding requirement (MFR). This was “a rigid and weak approach and resulted in some employers contributing the minimum possible, rather than making contributions appropriate to the specific circumstances of their scheme. In the most severe cases, this underfunding left members facing significant losses in the event of a scheme wind-up” (The Pension Protection Fund (PPF) and TPR, 2016, para. 24). Myners’ (Reference Myners2001) view was that the MFR regime was “distorting investment patterns without providing effective protection for members of defined benefit pension schemes” (para. 8.62). He recommended a “scheme-specific long-term approach based on transparency and disclosure, under which pension funds would report publicly on the current financial state of the fund and on future funding plans”.

A “scheme-specific funding” approach was introduced with the Pensions Act 2004; the PPF & TPR (2016, para. 26) explained:

The Act was also intended to comply with the EU Institutions for Occupational Retirement Provision (IORP) directive of June 2003, where section 15 referred to technical provisions in similar terms as the Act.

3.2. Discretion

There are potential benefits in recognising the specific circumstances of firms to influence the management of a scheme. Of particular importance is the strength of the employer covenant, that is, its ability and willingness to support the scheme. For example, a mismatched investment policy that leaves a scheme exposed to large changes in the difference between the values of assets and liabilities may be more acceptable if the firm has a strong covenant. Potentially relevant factors include the firm’s ability to pay dividends, the size of the scheme relative to the size of the firm and the options available to the firm to make changes in the scheme to cope with adverse situations (such as reducing benefits). Whether the firm has a need, in the short term, for cash to enable it to survive or develop would also be taken into account. On the other hand, if the firm has cash that is surplus to the requirements of the business, it may wish to pay higher contributions than otherwise. And while the employer covenant may be assessed with regard to the firm or firms that employ the scheme members, there is also a case for considering other firms in the group that may offer support in practice although not a legal obligation.

The different circumstances of firms are indeed important. The DWP (2017, para. 109) found considerable variation in deficit repair contributions as a percentage of pre-tax profit. So, as firms vary in their current financial position and future plans, there is merit in an approach that enables the actions taken to manage a scheme to reflect scheme- (including firm-) specific circumstances.

However, it is quite another matter for the assumptions of the actuarial valuation to be scheme-specific. TPR’s (2014) Code of Practice is explicit that there is discretion available in choosing a discount rate: trustees may use the flexibility available in setting the rate to reach a funding outcome that reflects a reasonable balance between the need to pay promised benefits and minimising any adverse impact on an employer’s sustainable growth.

The problem is that discretion can obscure the true finances of the scheme. It enables the reported financial position of the scheme to be manipulated, for example, by using less prudence in the assumptions, so that the liabilities are given a lower value and the resulting funding ratio is higher than otherwise. A deteriorating financial position may be hidden. The author believes that in order to manage a scheme effectively, it is helpful to have its financial position reported on a consistent, rather than a potentially distorted basis, to help enable scheme-specific actions to be identified and implemented. Where there is discretion on the valuation assumptions, it is “very easy to create the impression that… funding is progressing smoothly from one valuation to the next, even though the underlying solvency position may have changed enormously, which will obscure the risk from members” (Cowling et al., Reference Cowling, Gordon and Speed2005, page 82).

Further, the author believes that the information provided to members in the scheme funding statement is unhelpful. Consider members who are told that the funding ratio is 90%. However, if the discount rate was calculated with a greater degree of prudence, it could be, say 85%. Or if gilt yields were used, it could be, say 75%. It is difficult to see what reliance members can place on the ratio they are provided with. The scheme funding statement does not contain the basis for calculating technical provisions which, anyway, members would find difficult to understand. The discretion results in the funding information provided to members having little meaning when the calculation basis is adjusted to meet the needs of firms and trustees. With a defined valuation basis that removes discretion, member communication would be less open to manipulation and easier to understand.

In practice, is there variety in the discount rate? The answer is yes. The DWP (2017, para. 17.2) found “considerable variation in the discount rates used at scheme level”. The indicator used to assess this was the excess of the discount rate used (taken in “real” terms and converted to a single equivalent discount rate (SEDR) where more than one rate was used) over the 20-year real spot gilt yield. Since the discount rate reflects a (prudent) estimate of the expected investment return, this indicator is effectively a measure of the assumed outperformance of the scheme assets. The DWP (2017, para. 17.2) also found considerable changes in assumed outperformance over time. Indeed, 13.9% of schemes had a greater than 0.75% increase in assumed outperformance from one valuation to the next. Hence, it is not surprising that the government was “not convinced that there is strong evidence for a systemic issue with a lack of flexibility with the setting of the discount rate” (DWP, 2017, para. 17.1).

Evidence of variability in the discount rate (using the SEDR) is also available from data issued by TPR (2019) for valuations in 2016/17, see Table 1. Further, the variability is not random. This is suggested by TPR (2019) data: firms with a deficit and a relatively long recovery plan (10 years or more) used a discount rate of 2.892% on average, producing rather lower technical provisions than where the recovery period was under 5 years: 2.653% on average.

Table 1. Distribution of the outperformance of the nominal SEDR over 20-year spot rate on gilts. (Source: TPR (2019), Table 4.9a)

Evidence from valuations for the purpose of pension scheme accounts suggests that pension scheme assumptions can be chosen in a biased manner. When examining the assumptions to measure DB pension liabilities in UK FTSE350 firms’ accounts, Billings et al. (Reference Billings, O’Brien, Woods and Vencappa2017) found that weak schemes tended to use less prudent assumptions about price and salary inflation and the discount rate than strong schemes. Accounting rules restrict the discount rate to the yield on high-grade corporate bonds so that, while firms have some discretion on how to interpret this, that discretion is limited. It would be of interest to carry out similar research on funding valuations (although they are not publicly available in the same way as firms’ accounts). However, given that discretion in assumptions is greater in funding valuations than for valuations for the purpose of firms’ accounts, in particular as regards choice of discount rate, the author believes that it is reasonable to think that some bias will arise in choice of assumptions.

The Pension Schemes Bill (2020), as introduced in the House of Lords, proposes an offence of conduct risking accrued scheme benefits. Assessing whether an offence has taken place may be easier if there are valuations that do not involve the discretion on financial assumptions as currently available and which may lead to biased conclusions.

The concerns expressed above regarding discretion in valuation should not detract from the importance of scheme-specific management: rather, the contention is that clear and consistent valuation measures will provide better information to guide scheme-specific management.

7.1. Fulfilment or Transfer?

Should the liability be based on the liabilities being settled over time by the trustees (fulfilment basis) or by transfer to an insurer or other entity? The potential emergence of superfunds as a way to buy out DB scheme liabilities rather than through an insurer is a new development (DWP, 2018a). It remains to be seen how the pricing of superfund buyouts differs from an insurance solution; this paper focuses on the current practice of buyouts through an insurer.

The value of liabilities on the fulfilment basis is expected to be lower than using a transfer value, noting that an insurer:

• may make different assumptions regarding the level of interest, inflation and mortality (and other demographic assumptions that may be relevant);

• requires compensation to reward the risks (including investment risks) being taken;

• requires compensation for having to meet regulatory requirements, including maintaining a minimum level of capital;

• imposes a profit margin that may be more (or may be less) than warranted as compensation for risk, depending of the competitiveness of the buyout market; that margin to allow for tax as may apply;

• has costs of running the scheme, which may be higher (or lower) than those of the employer; and

• has transaction costs for implementing the transfer.

Funding is important because pensioners cannot rely on employers paying their pensions from revenue. The risk for scheme members is that the employer fails. If that happens, the benefits can be secured by a buyout. It is therefore important to carry out a valuation where the liabilities are assessed on a discontinuance basis, equal to the cost of transferring the obligations to an insurer. That is current practice in accordance with the regulations in force.

However, it is common for trustees, when disclosing the buyout cost in the summary funding statement, to say it is artificially high and not relevant when they are not planning a buyout. This reminds us of the development of IFRS 17, IASB’s standard for insurance contracts, when original proposals for the liabilities to be valued as the “exit” market value (“transfer basis”) were dropped in favour of a fulfilment basis on the grounds that insurance liabilities are typically fulfilled by the insurer and not transferred. Transfer is expected to inflate the liability value and would not be a neutral benchmark.

Solvency II for insurance is based around transferring liabilities to another firm, but that is specifically a requirement to reflect regulators’ wishes to provide for such a transfer if the insurer fails, whereas we are here seeking a value that is “neutral”.

The traditional calculation of a DB scheme funding ratio is based on the firm fulfilling its obligations to provide pensions and the cost of so doing.

Many DB schemes have transferred liabilities to an insurer. But, if a transfer has not been agreed, then if we wish to assess whether the assets are adequate for the liabilities, it is appropriate to work on the assumption that the firm will fulfil its obligations: a fulfilment basis is suitable for a benchmark valuation. We can then determine if the assets are on track to meet the liabilities.

7.2. Discount Rate: Matching or Budgeting Approach?

Discount rates are critical in pension valuations and it is a subject addressed by Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012). The background to this paper is of interest. The management board of the Institute and Faculty of Actuaries (IFoA) recognised that there were issues about discount rates in actuarial work and commissioned a report on the subject by Patel & Daykin (Reference Patel and Daykin2010). The board then appointed a group of senior actuaries with different backgrounds (insurance, pensions and investments) to review the report and prepare a paper which was presented at meetings of the actuarial profession and subsequently published in the British Actuarial Journal; it won the prize for the best paper by IFoA members in 2010/11.

The paper considers a valuation to assess the financial strength of a scheme assuming there is no other supporting financial entity; the valuation result can then inform what support is required. In this context, the authors concluded that a valuation to assess the assets required to meet liability cash flows should use a “matching” approach. Here, “the discount rates are consistent with the current market value of assets that, as far as possible, replicate the (future) economic behaviour of the liabilities to which the discount rate(s) might be applied” (page 170). This is consistent with the “law of one price”, where a set of cash flows has one price as, otherwise, an arbitrage opportunity arises. The authors also said, “Where objectivity and fairness between parties is paramount, and the assets and liabilities in question are essentially already contracted arrangements then a matching, i.e. market-consistent, approach, is nearly always likely to be preferable” (pages 175–176).

An alternative, “budgeting” approach, is where the discount rates are consistent with the expected returns on the assets expected to be held. According to Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012), this is suitable where the objective is advising on the accumulation of assets to meet the liability cash flows as they fall due (although it does not solve the problem of insufficient assets if the sponsor does not rectify a shortfall before adequate assets have been built up). Exley et al. (Reference Exley, Mehta and Smith1997) state that, strictly, such techniques “do not result in a liability valuation at all but rather a statement that a certain set of assets ‘100% fund’ a set of liabilities” (page 850).

Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012) applied their conclusions to measuring the solvency of DB pension schemes. They were critical of present practice where the emphasis is on the technical provisions, using a budgeting approach and a discount rate of the (prudent) expected rate of return: “from a member’s perspective this information is of limited use” (Cowling et al., Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012, page 179). Again, “member communication based on a budgeting calculation gives misleading information about the security of members’ benefits” (Cowling et al., Reference Cowling, Fisher, Powe, Sheth and Wright2017, page 60). Instead, Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012) concluded that pension scheme solvency should be estimated using a matching framework and, in disclosing the funding position to members, trustees and regulators, the focus should be on “the solvency position and how it is expected to develop under the agreed funding plan” (page 179). The author’s view is the same: it is not a communication issue as such – improving the communication of a result based on a budgeting basis is not the solution, which should be to communicate using a matching basis.

Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012) did refer to the possibility that some blending of matching and budgeting approaches may be appropriate, involving matching calculation constraints on the budgeting approach. This could be relevant to, for example, a DB pension scheme where some liabilities are more contractual and some less. The Institute and Faculty of Actuaries (2012) referred to the possibility of a blended approach in some situations. However, significant discretionary elements in DB schemes benefits are now uncommon, and the author feels that there is no reason for the solvency of a scheme to use some blended approach, the justification for which and operation in practice are not clear. Indeed, when Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012) applied their thinking to DB scheme solvency, it was clear that they regarded a matching approach as the one to take.

For a DB pension scheme, what are the matching assets? Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012) say that under a matching methodology the present value of a 10-year cash flow is calculated with reference to the market price of a 10-year bond. While this implies bonds as the match for pensions that are deferred or in course of payment (index-linked bonds as appropriate), aren’t equities a match for salary-related pensions? The analysis and modelling of Exley et al. (Reference Exley, Mehta and Smith1997) concluded that cash or bonds match fixed pension liabilities; inflation-linked bonds for inflation-linked labilities; for salary-related liabilities, equities were a poor match: “we would find it very difficult to justify any match between equities and salaries if index-linked bonds were available as an alternative” (page 898). Other authors have concluded similarly, for example, Chapman et al. (Reference Chapman, Gordon and Speed2001) concluded that the value of any cash flow stream is equivalent to the value of a matching asset: for DB pensions, bonds were the matching assets.

A further possibility is to use swaps, with a credit risk adjustment for bank default, as prescribed by Solvency II. Hatchett et al. (Reference Hatchett, Hurd and Clacher2013) suggested that, in the UK context, readers might think of risk-free assets as gilts or suitably well-collateralised swaps with a sufficiently reliable counterparty. Charmaille et al. (Reference Charmaille, Clarke, Harding, Hildebrand, Mckinlay, Rice and Reynolds2013) regarded the normal position to be for swap rates to exceed gilt yields, although this could be reversed in certain circumstances; Foroughi (Reference Foroughi2012) put forward possible explanations for negative swap spreads at medium to long durations. The PPF’s liabilities are referenced to the higher of gilt yields and swap rates. This is similar to Foroughi’s (Reference Foroughi2012) “minimum cost” valuation premise: where government bonds and swap asset classes are deep and liquid and form a good match for (insurance) liabilities, the more favourable result should be used. He put forward several reasons for preferring this “minimum cost” approach, including markets not always being perfectly efficient, and the benefit of reducing the risk that one calibration instrument is subject to high demand simply because of the market-consistent valuation. The author regards this as a suitable way ahead for DB schemes, although it is fair to admit that this may complicate communication of the valuation result.

The consensus of recent actuarial research has highlighted the case for solvency valuations using discounting at the yield on matching assets, with no recent papers suggesting that equities would be a good match for salary-related benefits:

• Exley et al. (Reference Exley, Mehta and Smith1997) argued that scheme solvency should be assessed with discounting based on the matched, not actual assets (i.e. cash and bonds);

• Head et al. (Reference Head, Adkins, Cairns, Corvesor, Cule, Exley, Johnson, Spain and Wise2000) found that an economic valuation using gilts is entirely objective and most satisfies (compared with other valuation types) targeting security as it targets the defined benefit in the event of scheme wind-up as well as at projected retirement age;

• Chapman et al. (Reference Chapman, Gordon and Speed2001) proposed discounting liabilities at bond rates to produce a market-consistent valuation;

• Cowling et al. (Reference Cowling, Gordon and Speed2005) recommended solvency be assessed by reference to a notional investment strategy with assets providing defined benefits with a high degree of certainty such as gilts, AAA debt or collateralised swaps;

• Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012)’s paper indicated that solvency measurement should use the discount rate on matching assets, which for DB liabilities were gilts;

• Hatchett et al. (Reference Hatchett, Hurd and Clacher2013) concluded that a valuation based on financial economics has several benefits, including providing a more objective measure of the pension promise; this would mean discounting at the yield on matching assets, that is, low credit risk assets such as index-linked gilts, insurance contracts and swaps; and

• Cowling et al. (Reference Cowling, Fisher, Powe, Sheth and Wright2017) stressed that a solvency valuation should use a matching framework, the matching assets being bonds or possibly other assets that generate a known series of cash flows.

Using discount “rates” rather than “rate” is important: account has to be taken of the term structure of interest rates (Exley et al., Reference Exley, Mehta and Smith1997).

It is useful to look at accounting standard-setters’ approaches, not to apply them to pension scheme valuations mechanically, as they have different objectives, but the conclusions and their rationale may suggest approaches that are helpful. The (UK) Accounting Standards Board (ASB), when developing a standard for pensions accounting in the late 1990s, proposed in its Discussion Paper that the matching assets were gilts for pensions that were deferred or in payment but equities for salary-related pensions. However, following the research of Exley et al. (Reference Exley, Mehta and Smith1997), it accepted that bonds, not equities, were the answer. When FRS 17 was issued in 2000, the discount rate was set as the rate on high-grade corporate bonds rather than gilts: this was because employers have an option to offer benefits lower than the best estimate of cash flows in the valuation of liabilities. For example, equity investing may lead to a surplus that is used to increase members’ pensions. Hence, the ASB felt that discounting using gilts would overstate liabilities. In an ad hoc approach, they decided to use the yield on high-grade corporate bonds. For the international standard, IAS 19, the IASB decided that the discount rate should reflect the time value of money and decided on high-quality corporate bonds (but, if no deep market in such bonds, government bonds). EFRAG (2008) suggested using a risk-free rate and the ASB confirmed its view that a risk-free rate should be used (EFRAG, 2009). It is the case in the UK that gilts would better reflect the time value of money or a risk-free rate than would corporate bonds. Indeed, to review the rationale of FRS 17, there is now little expectation that members will receive benefits above the minimum in scheme rules. The case for using risk-free yields is therefore strengthened. To illustrate that this type of approach is acceptable in accounting, it is used in IFRS 17 for insurance contracts.

It is therefore concluded that the adequacy of assets to meet DB pension liabilities should be measured with the liabilities discounted using risk-free yields. We would expect the UK pensions regulator to start considering gilts, including index-linked gilts, as an appropriate way forward. With respect to the argument that gilts are not risk-free, constructing a hypothetical rate curve deemed to be totally free of credit risk has its own difficulties, and Foroughi (Reference Foroughi2012) accepts “suitably low credit risk” as the way ahead. There are potential complications arising from some pensions having a minimum price indexation (0%) and a maximum; these can be addressed by deriving a hedge of gilts and index-linked gilts (Exley et al., Reference Exley, Mehta and Smith1997).

There are two areas where the gilt/swap approach needs adaptation, being areas of basis risk referred to by Charmaille et al. (Reference Charmaille, Clarke, Harding, Hildebrand, Mckinlay, Rice and Reynolds2013). One arises from the absence of assets with maturity terms over 50 years, also considered by Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012), who discussed possible ways of dealing with this. Second is the lack of CPI-indexed assets, an issue which the PPF monitors in its management reporting (Charmaille et al., Reference Charmaille, Clarke, Harding, Hildebrand, Mckinlay, Rice and Reynolds2013).

In contrast to market-based valuations, the common valuation practice, consistent with the regulations, uses a budgeting framework to calculate technical provisions, which takes advance credit for the equity risk premium. This does not, however, provide information on the solvency position of the scheme, which Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012) and (2017) regard as requiring a valuation using a matching framework. Matching involves considering a notional portfolio, so that the assets do not affect the liabilities, subject to secondary considerations such as stranded surplus and credit risk (Exley et al., Reference Exley, Mehta and Smith1997; Cowling et al., Reference Cowling, Fisher, Powe, Sheth and Wright2017).

To make the point with an example taken from Exley et al. (Reference Exley, Mehta and Smith1997), consider a firm that offers a guaranteed sum of £110 in a year’s time and invests £100 now in equities with an estimated return of 10%, when gilts yield 8%. Equities (where the return is not guaranteed) are not a match for a guaranteed liability. So, while the fund may end up with more or less than £110 and, on average, appears to break even, the fund has not received any compensation for the risk it is taking. Therefore, the value of £110 guaranteed a year ahead is not £100 now, even if equities expect to return 10%. Indeed, if shares in one firm X are expected to perform really well, with an expected return of 20%, could the firm use this to guarantee £120? (no).

Cowling et al. (Reference Cowling, Gordon and Speed2005) are concerned that using a discount rate higher than that derived from a matching portfolio is simply a way of reducing the contribution rate at the expense of higher contributions later and lower member security: “an element of self-delusion within the actuarial profession” they called it (page 78). Hatchett et al. (Reference Hatchett, Hurd and Clacher2013) accept that, while schemes may wish to take investment risks in order to reduce the cost to the firm, they have a concern that, under current regulations, stakeholders “could change the asset allocation to ‘hide’ the deficit” (page 292). The author would say that using a budgeting framework means that the liabilities appear to magically reduce if the firm decides to adopt a more risky investment strategy with higher expected returns, and that this is quite wrong: the liabilities depend on factors such as service, salaries and mortality. Indeed, as Hue et al. (Reference Hue, Junks, Spain, Bora and Siew2019) stated, “The ‘liabilities’ are really the estimated cashflows rather than the capital amount. Thus, the liabilities do not depend on the discount rate” (page 25). The author believes that, in contrast, a market-based view achieves greater objectivity, clarity and consistency.

It is fair to add that one paper, a report for the National Association of Pension Funds by Clacher & Moizer (Reference Clacher and Moizer2011), recommends the liabilities be discounted at a rate that reflects the scheme’s investment strategy, and this to be combined with a value of assets smoothed to reflect the long-term nature of the pension obligation. However, Clacher has clarified that this proposal is specifically in relation to company accounts rather than valuations for trustees, etc (see comments in the discussion on Hatchett et al., Reference Hatchett, Hurd and Clacher2013).

Recent years have seen interest in so-called “self-sufficiency”, with a survey by Mercer finding 47% of schemes are looking at this as their long-term objective (Dodds, Reference Dodds2018). Cowling et al. (Reference Cowling, Fisher, Powe, Sheth and Wright2017) suggest a number of possible ways to interpret what is meant by a self-sufficient scheme, typically implying a low level of reliance on the employer, although they add their view that a layman would interpret it as a scheme no longer reliant in any circumstances from the employer, implying that the assets should be sufficient for a buyout from an insurer. Some low mismatched investment risk may be regarded as optimal by trustees and firm, and a discount rate may be selected to equal the expected return on such a basis. However, as this is not a matching portfolio, the expected returns do not produce a discount rate to use to measure solvency in accordance with the matching principle.

Integrated risk management has been promoted by TPR (2018), which has said that the discount rate chosen should reflect risk management principles and be consistent with, inter alia, the view of the employer covenant, with a weak covenant being associated with a relatively low discount rate. That would lead to the liabilities being given a high value and could lead to a recovery plan and higher contributions than otherwise, arguably appropriate for an employer who may be unable to meet its obligations. However, Cowling et al. (Reference Cowling, Fisher, Powe, Sheth and Wright2017) found that the link between covenant and discount rate is, in practice, quite weak. Subsequent data from TPR (2019) suggest the link may no longer exist. In any event, we ought not expect the discount rate to be affected by the covenant: the liabilities depend on salaries, mortality, etc and are not affected by the employer’s covenant (aside from credit risk and stranded surplus). The author’s view is that scheme-specific management is best done with clear and consistent solvency valuations, and a weak covenant can in itself be persuasive of the need for a low-risk investment strategy and higher contributions without needing to adjust or benefitting from adjusting the valuation result.

7.3. Liquidity Adjustment or Not?

Another factor is in place: “Pension promises are typically illiquid, and hence the matching portfolio should also contain illiquid assets” (Chapman et al., Reference Chapman, Gordon and Speed2001, page 618). Charmaille et al. (Reference Charmaille, Clarke, Harding, Hildebrand, Mckinlay, Rice and Reynolds2013) mention this as a possible future development in managing the PPF.

Liquidity adjustments have been the subject of much debate, especially for insurers with a portfolio of annuities. Keating et al. (Reference Keating, Hatchett, Smith, Walton and Zhao2016) described liquidity as “the degree of inter-exchangeability of money and the asset” (page 7). However, there are some tricky issues. We think of liquidity as an attribute of assets: how easily can they be traded and with what impact on price. How does liquidity relate to liabilities, when they are typically not traded? Foroughi (Reference Foroughi2012) refers to different ideas on a liquidity adjustment: either using

• a concept of a hypothetical asset that matches the characteristics of the liability outflows, including liquidity or illiquidity; or

• derived from the return on achievable viable asset-liability management strategies.

We can consider an annuity as an asset itself in the hands of the policyholder who, unable to trade the annuity requires compensation for this illiquidity (Browne et al., Reference Browne, Milevsky and Salisbury2003). However, this does not translate into how we might adjust the valuation of an insurer’s (or DB scheme’s) annuity liability. There are a number of approaches to isolate what is often referred to as the illiquidity premium (in terms of the yield on assets), though one could also start from the basis that liquidity has a cost (Keating et al., Reference Keating, Hatchett, Smith, Walton and Zhao2016) and hence there is an illiquidity discount, that is, the value of illiquid assets (and liabilities) is below the value of the corresponding liquid cash flows.

In Solvency II insurers may, in specified circumstances, use a “matching adjustment”. Here (and this is to simplify a complex process), insurers can discount future predictable cash flow obligations at the usual risk-free rate plus the excess of the yield on “matching assets” they have over the sum of the risk-free rate and a “fundamental spread”. That spread reflects the compensation for risks to which the insurer is exposed, dependent on the likelihood of default; it is argued that, on corporate bonds, the excess reflects the compensation, or “liquidity premium” for the risk that insurers may have to sell into secondary bond markets when credit spreads are wide (Rule, Reference Rule2018). Annuity providers with matching assets are not exposed to this risk and can therefore earn the liquidity premium if they can bear the credit risk and hold the assets until maturity (Rule, Reference Rule2018).

The average matching adjustment for corporate bonds was around 120 basis points at 31 December 2016, although the benefit was even greater for illiquid assets such as infrastructure loans or equity release mortgages, where the matching adjustment could support an increase in the discount rate of about 150 or 200 basis points, respectively (Rule, Reference Rule2018). These might be termed “illiquidity adjustments” (IAs), reflecting the additional yield from holding assets to maturity and avoiding the risk of early cashing in. They are consistent with Dullaway’s (Reference Dullaway2009) argument that there are times when there are multiple ways of constructing replicating portfolios that are essentially risk-free and which match annuity portfolios, in which case it is reasonable for insurers to use the cheapest way to value annuity business. Similar is the argument that the discount rate in Foroughi’s (Reference Foroughi2012) “minimum cost valuation” need not be restricted to government bonds and swaps as other assets may be able to match insurance liabilities. In principle, arbitraging opportunities should limit the possibilities, so this may turn out to be not material. The author’s view is that the use by DB schemes of illiquid assets suggests that, if the allowance for credit risk in their yield is removed, they could, in principle, be incorporated in a minimum cost valuation.

The Solvency II matching adjustment is available only if the insurer holds the assets concerned, and one possibility for pensions regulators is to allow an IA that is supported by assets the scheme holds. However:

• this is contrary to the principle of a matching framework as suggested: it would mean that two schemes with identical obligations could show different liability values depending on the assets they had (breaching the principle discussed earlier that the valuation of liabilities is not materially affected by the asset mix);

• there may be an undue incentive to move into assets that can lead to an IA credit, especially if the assets are complex to understand and manage, and a pensions regulator may have concerns about pension schemes being led to take more risks with illiquid assets, especially as regulatory oversight on pension schemes is less intrusive compared with that for insurers under Solvency II; and

• it could be administratively complex and costly, especially if an approach similar to Solvency II is taken.

If an IA is available independent of the assets a scheme holds, a potential issue is that if a scheme holds gilts and swaps equal in value to the liabilities after discounting using an IA, we would not expect the liabilities to be met in full. Trustees need to be clear on whether their policy is to choose assets that earn an extra return as a reward for holding to maturity. This reinforces the need for schemes to document their approach to financial management.

Should the discount rate then use the highest possible IA? This highlights the situation where we have different yields apparently “as if risk-free”, having accounted for the fundamental spread. It is therefore reasonable to suppose that some (especially the higher) IAs will reduce over time (see Rule, Reference Rule2018) and indeed pension funds have found that increased competition for a higher rate of return is leading to the reward for illiquidity declining (Government Actuary’s Department, 2018).

Other factors suggest caution in using IAs. There are several ways to estimate the extent to which yields reflect risks not borne by the entity holding the assets to maturity, each method having issues (Telford et al., Reference Telford, Browne, Collinge, Fulcher, Johnson, Little, Lu, Nurse, Smith and Zhang2011; Foroughi, Reference Foroughi2012). Care is needed in assessing the fundamental spread with illiquid assets that are not rated (Rule, Reference Rule2018). Further, since DB schemes offer options, including transfers to other schemes and cash commutation at retirement, it would be less easy to apply an IA for pre-retirement liabilities. Indeed, Duval (Reference Duval2019) commented that “most managers these days will tell you the illiquidity premium is zero or possibly negative” (page 5). On this basis, it would be fair to accept a pensions actuary concluding that using an IA would be inappropriate or immaterial.

It could be argued that a scheme with projected positive cash flow for some years is not subject to significant risk of shortfalls arising from early encashment of assets, this justifying an IA of greater size than otherwise. It is fair to say that an open scheme is more able to invest in illiquid assets without having to sell them to pay pensions (Salt, Reference Salt2019). That may well lead to a higher IA. However, there is a need to consider the possibility of the scheme ceasing to admit new entrants and/or closing.

IFRS 17 for insurance contracts requires the discount rate to reflect, inter alia, the liquidity characteristics of the liabilities. It is not yet clear how this will be interpreted and implemented. Since life insurers will need to consider their annuity portfolios, it could be useful for life and pensions actuaries to work together in researching the issues of illiquidity.

Pensions regulators may conclude that they should not allow any IA as theory and research as incomplete, but this could produce an upward bias in the calculation of liabilities.

One possibility is an approach whereby schemes discount at a risk-free rate (higher of gilts and credit risk-adjusted swaps) plus a factor that reflects some IA that would be available, independent of the assets actually held, but as if the scheme used assets such as corporate bonds and illiquid assets for part of its portfolio; this either only applied or only applied fully to the post-retirement liabilities. In principle, the IA varies by scheme characteristics and will change over time. TPR may wish to question how an actuary determines an IA that has been used, and it would be hoped that experience would build up over time.

7.4. Allow for Credit Risk or Not?

Since the objective of the valuation is to ascertain how members’ entitlements compare with the assets of the scheme, should the liabilities be reduced to reflect the credit risk of the firm rather than be calculated at a risk-free rate? Current practice both in valuations for trustees and for the accounts is not to allow for the firm’s credit risk; the basis for conclusions for IAS 19 explicitly rejected allowing for credit risk on the basis that a weak firm would then show a lower level of pension liabilities.

Ralfe et al. (Reference Ralfe2018) put forward an alternative whereby the discount rate is the market rate on bonds with a similar obligor to the employer, taking account of the security provided by the fund assets. This allows for credit risk. It may well be a useful calculation from the perspective of the employer in assessing the expected burden of its pension obligations. There can be some practical difficulties. In particular, if the firm has not borrowed on the market, the required discount rate is not readily observable; to discount pension obligations we need rates at different durations, including durations longer than the typical length of corporate borrowing; and the discount rate needs to take account of the assets already in the fund.

The actuarial literature supports solvency valuations making no allowance for credit risk:

The profession’s discount rate project came to the same conclusion: a solvency valuation should compare the assets with the value of commitments to members, which requires no allowance for credit risk (Cowling et al., Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012). Such allowance is a separate, albeit related, issue (Cowling et al., Reference Cowling, Gordon and Speed2005). It may be good practice for actuarial valuations to also calculate a figure for liabilities with allowance for credit risk. Indeed, the solvency figure on its own is just one figure and does not tell us the whole story.

The author’s conclusion is that allowing for credit risk is inappropriate for a benchmark valuation on solvency as it is more geared to the needs of the firm (although still relevant for members) and requires more assumptions than a valuation with no allowance.

7.5. Best Estimate or Not?

The valuation of liabilities requires assumptions about several matters which affect future cash flows: mortality in particular, and other demographic factors such as the proportion of members who are married; member options, including cash commutation; and expenses. Since we are using a matching approach and aiming for a valuation which is neutral between members and employer, the cash flows should be at the expected value. Given the uncertainty involved, it is appropriate to use probability-weighted outcomes. This is consistent with IFRS 17 and Solvency II for insurance contracts.

The author regards this as preferable to using the term “best estimate” (as in IAS 19), which is open to differing interpretations.

In some cases, the probability-weighted expectation may be approximated by a single future assumption, for example, one mortality table instead of a series of future mortality assumptions weighted with probabilities. This recognises the difficulty of attaching probabilities to a range of uncertain outcomes, often some long time in the future.

While regulations would prescribe the use of expected values, scheme-specific information on demographic and similar variables, notably mortality, would be considered in determining the precise assumptions to use.

7.6. Prudent or Not?

Should the assumptions be prudent assumptions? Since we are seeking an unbiased measure of liabilities, we wish to know how best they can be represented rather than err on the high side. Hence, if prudence means assumptions which are biased in favour of overstating the liabilities, then the answer is no. As Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012) stated in the context of the discount rate, “matching valuations contain no concept of a prudent valuation: the market value of a matching asset portfolio is a single number, not a distribution” (page 208). Mortality, expense, option and similar assumptions could be prudent, and Cowling et al. (Reference Cowling, Gordon and Speed2005) indicated that this should not be controversial for a solvency measure; indeed, they asserted that negative options available to parties not running the scheme should be allowed for fully. Indeed, following the IORP Directive, assumptions in technical provisions in pensions have to be chosen prudently, taking into account the firm covenant. However, prudence in the sense of producing a liability value higher than the expectation produces a biased estimate, which the author argues is inappropriate for the benchmark.

It is useful to add the thinking of the International Accounting Standards Board (2018) in saying that prudence is the exercise of caution when making judgements under conditions of uncertainty but does not mean that assets or liabilities should be either overstated or understated. This view of prudence is useful in interpreting the IORP Directive.

It is also noted that IRFS 17 and Solvency II for insurance do not add margins as part of the liabilities, specifically for prudence.

Prudence is also a problem as there is a divergence of answers as to what constitutes prudence and no right answer (Cowling et al., Reference Cowling, Gordon and Speed2005). And prudence may also lead to difficulties regarding transparency. The reader of the resulting report may, even with extensive disclosure, find it difficult to understand what prudence has been incorporated and what its effect is. For example, if it is stated that, to be prudent, mortality rates have been assumed to decrease at ½% p.a. more than expected, what does the reader make of it? The author regards it as more appropriate to use expected values for assumptions and then have risk disclosures that illustrate the effect of variations on those assumptions. A consistent series of valuations using expectations is argued to be clearer than one where prudence is built in and may be varied from one valuation to the next, thereby hiding what has been happening. Prudence may produce unwarranted complacency or may provoke undesirable consequences if pension obligations are thought to be more onerous than they really are.

It is very welcome to have prudence in the management of pension liabilities: trustees must understand the obligations to and expectations of members and acknowledge the implications of the scheme for the employer who has a business to run. There is a need to balance different interests and avoid reckless risk-taking. But prudence in management does not mean there has to be prudence in accounting, if that were to mean inflating the level of liabilities beyond what is derived from using expectations.

7.7. Risk Margin or Not?

As an employer with a DB scheme has committed itself to a liability that is uncertain, it needs to ensure it can meet its obligations even in the event of adverse outcomes when assumptions turn out to differ from expectations. Hence, it can be argued that a risk margin should be added to the expected value of the liability.

To begin reviewing ideas on the subject, IFRS 17 requires the liability value on insurance contracts to include a risk margin, meant to be what the insurer would pay to have a certain rather than uncertain liability with the same expected value. If we assume that the insurer is seeking to maximise shareholder value, then the risk margin is the effect on shareholder value of the uncertainty. Such a risk margin is surprising, as firms’ financial statements do not account for shareholder value. In any event, this would not be suitable for a benchmark valuation of a DB scheme: reflecting shareholder preferences would mean it is a biased rather than a neutral figure, aside from the severe difficulties in calculation.

Value at Risk (VaR) and Tail VaR have been put forward as possibilities for the risk margin under IFRS 17 (see International Actuarial Association, 2018), and while they do not provide a link with shareholder value in accordance with IFRS 17, this does not disqualify them as candidates for a risk margin for a DB scheme. Cowling et al. (Reference Cowling, Gordon and Speed2005) suggested a VaR measure for margins for risks that cannot be matched. If a VaR or similar approach is to be pursued, there are issues of what percentile in the probability distribution to use, together with the practical problems of establishing the probability distribution of not only mortality but other demographic variables, expenses and options as well. Noting that options such as cash commutation are subject to change if regulations change exemplifies the problems. Communication of the result may also be a challenge: a comment such as a 95% risk level needs to be understood to cover risks that are not investment risks (which, in practice, are likely to be the main risks).

Difficulties of quantification were the reason that EFRAG (2008), when considering pension liabilities in a firm’s accounts, concluded that although liabilities should reflect a risk margin, a better approach is to disclose the sensitivities of the liabilities to changes in the assumptions. Similarly, Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012) indicated that a series of stress tests was a way forward.

The PPF looks to have 90% confidence of meeting unexpected claims over 5 years and longevity risk over the remaining lifetime of the fund and estimates that a margin of 10% of liabilities is required for this (Charmaille et al., Reference Charmaille, Clarke, Harding, Hildebrand, Mckinlay, Rice and Reynolds2013). The 10% is not fixed but will be re-evaluated as conditions change. Other schemes could make calculations to derive figures suitable for their circumstances. One possibility is for pensions regulators to impose some similar type of risk margin for non-investment risk unless a scheme demonstrated that it was inappropriate in its case.

Now consider another possible approach to the risk margin: the amount needed, above the liabilities, to enable another firm to take over the liabilities. This is the rationale of the risk margin in Solvency II; we need not here be concerned with the controversies around the calculation method prescribed. In a DB context this is the amount that, when added to the liabilities, gives the buyout cost.

The importance of the buyout cost is illustrated by it already forming part of the existing communication to members. Cowling et al. (Reference Cowling, Frankland, Halls, Kemp, Loseby, Orr and Smith2012, page 178) say that “Regulators and others are increasingly driving the purpose of funding a pension liability to be to provide security for members’ benefits in the event of the employer becoming insolvent”. Indeed, if the pension assets are there to provide security if the employer is no longer there, then it is what happens in that event that is important. If the firm is insolvent and the result is a buyout, then it is the buyout cost that matters.

Using the buyout cost as the risk measure is relatively straightforward given that it is already calculated as part of the valuation process, although usually an approximation based on the actuary’s knowledge of the market rather than a response to a specific quotation request. It avoids the problems of a calculation involving VaR, etc or the effect of risk on shareholder value and appears to be a reasonable approach for determining the risk margin for a DB scheme.

Should the risk margin form part of the liabilities as reported? This would lead to the liabilities being reported at a higher a level than was expected to arise. This is inconsistent with the requirement set that the benchmark valuation be unbiased, and the author therefore concludes that the answer is “no”. Further, if the risk margin were as suggested above, this would lead to the liability being the buyout cost, and we know from current practice that trustees could suggest that scheme members pay little regard to it.

However, trustees may rightly be concerned at the prospect of not having enough assets in their fund in the event of adverse experience. It is understandable that they may wish to retain some or the whole of the risk margin in the fund to ensure they have assets under their control if risks turn out badly, in the sense of leading to the solvency position being less favourable than projected. On the other hand, a firm may argue that it is an inefficient use of its resources to fund against unexpected risks, which may not arise and if they do arise, may be offset by other risks turning out better than expected. It would also increase the possibility of stranded surplus.

There is merit, from the perspective of trustees and members, of a risk margin being included in the fund. There are a number of possibilities as to what trustees should aim for (VaR, Tail VaR, stress tests, etc). The measures of risk available, covering both investment and other risks, can inform discussion between trustees and firm on what funding for a risk margin could be acceptable.

The author does not propose a regulatory measure for a risk margin to be held in the fund. This reflects

• there is no clear objective answer for a risk margin; what is optimal depends on the needs of trustees/members and the firm;

• rejection of proposals for an amended IORP Directive to include a specific amount of capital requirement;

• regulators and trustees will have access to risk disclosures, which can inform discussions on funding; and

• the need for further research on liability measures currently used and as proposed here and the quantitative impact of possible risk margin measures.

Hence, the conclusion is that a risk margin should not be included in the reported liabilities; while employer and trustees may agree some risk margin to be held in the fund, this would not be required by regulations; and disclosure of the buyout cost should continue.

7.8. Conclusion

The benchmark liabilities would, as part of a solvency valuation, be prescribed as:

• the benefits on discontinuance;

• discounted using the greater of gilt yields and swap rates;

• with, in principle, an “illiquidity adjustment” reflecting the way an annuity is illiquid, allowing schemes to hold assets without exposure to the costs arising on early sale and earning a yield above that on gilts/swaps, this adjustment being independent of the assets held; and

• a probability-weighted expected set of assumptions for mortality, etc.

The valuation process would also estimate the buyout cost.

However, a valuation with just two numbers, assets and liabilities, is insufficient. Risk is important, even if a scheme is closely matched on investment strategy; 100% funded on a solvency basis now does not necessarily mean 100% funded in the future. Section 7.9 refers to risk disclosure, including sensitivities.

7.9. Risk Disclosure

The valuation needs to be accompanied by risk disclosure, taking account of risks from non-matched assets as well as non-financial risks. One way is to use sensitivity tests, covering both assets and liabilities, the latter encompassing not only financial but also other risks, in particular demographic risks. A simple possibility is to start using the method (or something similar) of calculating “stressed” assets and liabilities in determining the levies that schemes pay (Pension Protection Fund, 2019), for example, UK equity prices falling by 19% and nominal interest rates reducing by 0.75 percentage points. The DWP (2018a) suggested TPR could issue a code of conduct to include stress tests and sensitivity analysis for superfunds, and such an approach could be used for DB schemes in general, and may go beyond what is currently included in some DB scheme actuarial valuation reports.

An alternative is a stochastic analysis, which can encompass variability in both assets and liabilities and give a full range of probabilistic outcomes. Cowling et al. (Reference Cowling, Fisher, Powe, Sheth and Wright2017) expressed a preference for using asset models with economic scenario generators to determine confidence levels. Models can be valuable in indicating risks and in identifying the scenarios that produce the most adverse results, enabling decisions to be taken on whether risk mitigation is needed. The DWP (2018a) has suggested stochastic modelling as one possibility as part of a new regulatory framework for superfunds.

Important issues with stochastic analysis are model risk, highlighted with the global financial crisis of 2008; model manipulation (Mariathasan & Merrouche, Reference Mariathasan and Merrouche2014); the cost of undertaking the work, which may not be warranted for small schemes; limited evidence on the probabilities of some assumptions and the challenges in communicating the outputs, assumptions and uncertainties.

Pensions regulators may nevertheless wish to use, in their monitoring work, risk measures from stochastic analyses that schemes produce and to require such analysis from schemes if not otherwise available.