Quantifying sponsor covenant risk for defined benefit pension schemes

Transcription

1 Quantifying sponsor covenant risk for defined benefit pension schemes Gazelle has developed a proprietary covenant risk quantification model MT (Mousetrap) which simulates scheme funding outcomes over the life of defined benefit schemes and calculates a monetary value for sponsor covenant risk. Objectives The concept of covenant risk encapsulates the exposure to loss for a defined benefit pension scheme due to the uncertain ability of the corporate sponsor to make its obligated contribution payments set out in the employer covenant. This paper aims to outline a quantitative approach and methodology required to specify a monetary value for this risk. Current practice in the UK pensions market The UK Pension Act 2004 established the Pensions Regulator (tpr), which has encouraged pension schemes to commission regularly an independent assessment of covenant risk. Covenant risk assessments have become central to establishing whether there is a detriment suffered by schemes following corporate transactions. There is a growing conviction that covenant risk should be reflected in the formal actuarial valuations of schemes, particularly Triennial Valuations, and individual scheme actuaries now regularly take into account covenant assessments when setting the Technical Provisions discount rate for scheme valuations. tpr guidance notes highlight those areas of investigation that are now expected best-practice for covenant assessment and include two specific areas for financial analysis : (i) the affordability of pension contributions and (ii) the formal position of the scheme on insolvency with an estimated recovery for the scheme. In addition there has been increasing recent emphasis on examining the impact of investment underperformance on covenant risk and in exploring the relationship between covenant risk and investment risk. This has given rise to a wide variety of covenant assessment methods with different approaches to judging whether a corporate sponsor s covenant is strong or weak. However, there is currently no objective, consistent and widely accepted framework for quantifying covenant risk. Generating a monetary value is valuable, since it can be used by scheme actuaries on a market-consistent basis to adjust the Technical Provisions discount rate, and thus objectively reflect covenant risk. The current inability to quantify covenant risk also poses serious issues for assessing detriment arising from corporate transactions where a before-and-after comparison can be difficult to make in qualitative terms with the objectivity necessary to provide agreement on mitigation. A quantitative basis for comparing investment risk and covenant risk is needed to enable trustees to consider how much investment risk is prudent for their particular situation and scheme. Finally there is a serious question concerning the limitations of current Asset Liability Modelling (ALM) because these models currently treat contributions as determined or given, whereas in reality they are uncertain, just as investment performance is; in fact, the sponsor covenant risk to which contribution payments are exposed is arguably less manageable than investment risk. Default risk If a corporate sponsor reaches a point at which there is no access to further funding for its business, it follows that after this point there will be no money to continue paying future pension contributions. The

2 period approaching termination of funding can take many twists and turns and the actual legal expression of termination can take several jurisdictional forms: typically insolvency or administration in the UK. The approach taken to dealing with an end to funding in debt and credit markets is to define a default event to include specified jurisdictional categories of corporate funding termination, then statistically estimate (or devise a model to predict) the expected frequency or probability of the default event. This is commonly termed default risk. A fundamental premise of this paper is that covenant risk is the same concept as default risk, applied to the pension creditor rather than to debt or bond creditors. Until default of the corporate sponsor, the prospect of continuing pension scheme contributions remains. At default a financial outcome is crystallised which represents finality in terms of future pension scheme funding. If covenant risk is the same concept as default risk then the quantification, risk measurement techniques and monetisation deployed by credit and debt market participants and their financial regulators should be just as relevant to quantifying and monetising covenant risk. This paper explores the differences and peculiarities of the pension creditor compared to the holder of debt or bonds. These present certain challenges, meaning that existing credit and debt market risk measurement techniques need to be adapted and extended for use with pension schemes in order to quantify covenant risk. The paper outlines how this can be done. Risk measurement techniques available from debt markets The methodology for quantifying default risk is well developed in debt and credit markets and is widely used by banks and bank regulators as a principal tool of risk management and in determining capital adequacy for lending institutions. The starting point for quantitative credit analysis is to design a methodology for a ratings system which will effectively discriminate between corporate credit risks (i.e. stronger credits from weaker ones), and provide a relative ranking of corporate credits. The ratings industry has developed over the last 30 years based on successful risk discrimination and by developing datasets that validate the effectiveness of their risk discrimination methodologies. These are well known and understood by debt market participants. Externally-generated credit ratings are widely available for many larger companies worldwide or can be obtained from models developed by rating agencies to generate ratings for unrated companies. Most banks and lending institutions have developed their own rating methodologies and systems (internal rating systems or IRS) to discriminate between credit risks they experience in their own particular areas of focus within debt markets. The Basel Committee on Banking Supervision has paid particular attention to testing and validating IRS. The ratings agencies hold extensive datasets of historical default rate data from which historical default rate frequencies can be derived for each credit rating category. The key building block for quantitative risk assessment is, for each credit rating category, the marginal probability of default ( PD ) in each time period for which data are collected (typically in one year intervals). The PD is the default frequency for a particular year in a times series (e.g. the probability of the corporate defaulting in year 4 of a loan) for credits in the same rating category. Adding the annual marginal PDs in each year of a time series (typically for the term of the loan or bond) gives a cumulative default rate over time which can be plotted as a default curve. This shows the development of the probability of default over time based on extensive datasets of actual default data. Importantly the historical default rate data indicate that the marginal probabilities of default are not linear throughout a time series, which results in a default curve and not a straight line extrapolation. This approach provides lenders with a quantitative tool to assess the probability of default for any given term of loan and borrower. The component default curve cumulative PDs are readily available as published data from ratings agencies such as Standard & Poor s or Moody s. Considerable work has been directed at explaining the shape of default curves for different rating categories and at designing predictive models. For the purposes of this paper the view has been taken that default curves are best derived from evidence-based historical default rate data. The 2 P a g e

3 search for a predictive model which can explain the observed data and predict future default rates is not the concern of this paper. The PD time series for each rating category provides the basis from which to quantify the default risk on a particular loan. This is undertaken by estimating the Loss Given Default ( LGD ) of the loan. The LGD is a financial estimate of the expected amount of the loan that would be recovered on a default event. The rating agencies helpfully provide data on loan recovery rates over time and for different industry sectors. Most lending institutions will undertake a detailed recovery assessment on a discounted cash flow basis incorporating estimated liquidation values for different collateral, recovery time periods and costs of recovery. To be precise the LGD for loans should be calculated for each time interval reflecting the debt repayment schedule over the term of the loan since loan repayments should, other things being equal, reduce the LGD as the loan approaches its term. For each time interval the marginal PD is then applied to the relevant LGD for that time interval to value the risk exposure ( RE ) arising. The expected loan risk exposure at default ( EAD ) is then the sum of the REs for each time interval comprising the term of the loan. For bonds without repayment schedules the cumulative PDs are applied to the estimated LGD for the time period under consideration (usually to maturity of the bond). Applying this approach to quantifying default risk across a portfolio of loans gives lending institutions the best available monetary estimate of their loan default exposures and therefore the level of capital they need, or are required by regulators, to maintain against a given loan portfolio. Relevance of debt risk measurement techniques for pension schemes A methodology for quantifying covenant risk is not currently available for use by pension schemes, in contrast to its widespread acceptance and use in debt and credit markets. This may seem surprising given that debt instruments and pension creditors share a common mutual dependency on the financial wellbeing of the corporate issuer or sponsor. This common dependency importantly represents a shared exposure to the same default risk. In principle the pension creditor should be as keen as lending institutions to quantify its default risk. The most likely explanation for the current lack of an approach and methodology for quantifying covenant risk lies in the way in which pension obligations have gradually been transformed by successive legislation from best endeavours intentions to hard commitments. Furthermore the differences in structure between pension deficit obligations (pension credits) and loans or bonds create additional complexities in the case of pension credits which need to be incorporated and resolved by any methodology capable of gaining practical and widespread acceptance by pension s professionals. One important difference between the structure of pension credits and loans or bonds concerns the length of the expected term exposure. The ratings industry is primarily focused on the banking industry and loans with terms of typically under 15 years duration whereas the material pension credit exposure might typically be and potentially as long as 70 years. This in turn gives rise to a potential concern regarding transferring debt market risk measurement techniques into the pension space: namely, that because default data is largely published for the banking audience with a 0-10 year fixed term, the underlying datasets do not extend beyond a 10- year period and therefore the PDs can t be accessed to give longer-term default curves. In practice however, external rating agency data is available for at least 20-year time horizons and much care is taken by ratings agencies to ensure statistical accuracy and minimise bias for longer horizons for example, by adjusting for rating withdrawals. In fact, various Moody s studies have shown that precision of default rate estimates increases with the length of the horizon 1. Extrapolating these datasets beyond 20 years by using, for example, moving averages is a practical approach given the very limited impact of PDs this far out in time. 1 See for example Moody s Special Comment, April Found at 3 P a g e

4 In addition, it is always open to expert covenant assessors to put forward other rating methodologies (such as an Internal Rating System or IRS) if they believe them to have superior discriminatory power for assessing corporate credits over longer time horizons. At this juncture however it may be difficult to back-test these IRSs as they will lack the extensive datasets built up over many years by ratings agencies. This paper therefore concludes that default curves, which are widely accepted and used in debt and credit markets, are both available and valid as the only current evidence-based methodology for use in quantifying covenant risk. Arguments between rating agencies over the best datasets to use in quantifying covenant risk are not the concern of this paper. The analogue of LGD for pension schemes can already be derived from the types of insolvency analysis undertaken by covenant assessors in line with tpr guidance. Based on current market practice it is arguable whether pension scheme LGDs are yet being assessed with the level of rigour and analysis undertaken by lending institutions, but it is difficult to see why the approach should be any different for either debt or pension credits, other than to ensure that the formal legal covenant support structure of participating employer companies provides the context for assessing pension scheme LGDs. The legal covenant support structure may well be very different to the debt financing structure for the same corporate and this fact, combined with differences in the priority of claims, may naturally result in markedly different LGDs for debt holders and pension creditors for the same corporate credit. A final consideration in assessing the LGD of pension schemes is the relevant regulatory framework. In the UK the PPF is not funded as a guarantee fund but it is generally viewed as inconceivable that it would not be supported by HMG. It can be argued that this effectively creates sovereign risk status for pension scheme LGDs up to 90% of accrued pension scheme benefits. While this is again clearly a significant difference compared to corporate loans and bonds it can, if required, be factored into the LGD financial analysis. Determining the term structure of pension credits Two key structural features of pension credits which are different to loans and bonds are that they are neither fixed in monetary amount nor fixed in term. These differences help explain why risk quantification techniques available in debt markets have not yet readily transferred to the pensions space. The structure of pension credits in terms of monetary amount and term derives from actuarial modelling of a pension scheme using scheme-specific actuarial assumptions regarding longevity, inflation and investment performance inter alia. Scheme-specific actuarial models assume pension contributions from the corporate sponsor namely the recovery plan contributions schedule for schemes that are closed to future accrual and normal contributions for schemes that are not closed to future accrual. Dependency on the corporate sponsor for future contributions can normally only be broken by two types of event. One is a default event as discussed above, while the other is the pension scheme achieving funding self-sufficiency. Funding self-sufficiency is achieved when the scheme s assets are deemed capable of funding the scheme s liabilities without need for further corporate sponsor pension contributions. There is also a third option for breaking dependency on a corporate sponsor which is a buyout of the scheme; this effectively provides additional capital or capital support for the pension scheme to enable it to become self-sufficient in terms of funding and no longer dependent on a corporate sponsor. This paper asserts that the actuarial calculation of the expected point in time at which a pension scheme achieves funding self-sufficiency, based on scheme-specific actuarial assumptions, is central to determining the quasi-term structure of a pension credit and therefore the timescale over which the pension scheme is exposed to default risk of the corporate sponsor. Once the term of the pension credit is known, exposure to default events is determined for the requisite number of time intervals through PDs and LGDs, giving a monetary value to the pension scheme s covenant risk exposure. The term in years to funding self-sufficiency also assumes a set schedule of pension contributions which can be used, along with other input data, to calculate pension scheme LGDs for each time interval or 4 P a g e

5 year. It should be noted that, in the absence of a secondary funding objective, self-sufficiency will not be achieved until the last liabilities of the scheme are discharged many years hence. The actuarial concept of funding self-sufficiency is therefore the important bridge to facilitating the transfer of risk quantification techniques from debt and credit markets to the pension space. It allows pension schemes to read off where on the default curve they are when dependency on the corporate sponsor ceases and therefore defines the time limit on exposure to default risk. Therefore, with the appropriate adaptation and reliance on the actuarial construct of scheme selfsufficiency, the techniques of risk quantification available to debt markets can be successfully transferred to the pension space. Importantly the approach being developed is not dependent on any particular chosen definition of scheme self-sufficiency but rather requires that a sensible definition is chosen which generates an expected point in time at which scheme self-sufficiency is achieved. Features of pension credits requiring additional developments in risk quantification A further fundamental issue results from the use of the self-sufficiency concept as the determinant of pension credit term, because the point in time at which scheme funding self-sufficiency is expected to be achieved is itself a dynamic construct. In essence the point in time at which a pension scheme reaches self-sufficiency is a function of three principal drivers of the actuarial financial model. These are: (i) (ii) (iii) An assumed rate of pension contributions; An assumed level of investment performance from scheme assets; and The timing of the scheme s benefit cash flow payments reflecting inter alia the expected longevity of scheme members These are potentially exciting links to explore because they suggest that the methodology for quantification of covenant risk will reflect in turn: (a) (b) (c) The affordability of an assumed rate of pension contributions by the corporate sponsor; The assessed risk of investment under- or over-performance by the scheme s assets; and Key assumptions such as longevity risk which might expand or contract the scheme s liabilities Constraints or adverse variations applied to (a), (b) and/or (c) will have the conceptual impact of pushing out in time the point at which scheme self-sufficiency is achieved, thereby causing a righthanded drift along the default curve which then exposes the scheme to increased covenant risk. The most exciting aspect of this is that these relations provide a basis and quantitative framework for an integrated approach to pension risk assessment, quantification and management which joins up covenant risk, investment risk and liability risk. Affordability of pension contributions and covenant risk tpr guidance directs covenant assessors to examine the affordability of pension contributions by a corporate sponsor. There is however currently no framework or methodology that provides a basis for quantifying the contribution or importance that affordability makes to covenant risk. We have put forward the argument above that covenant risk is no more than default risk measured from the perspective of a pension creditor. The question that needs addressing now is the relationship between affordability and corporate sponsor default risk. Affordability is a financial concept designed to measure the ability of one party to pay in full its financial commitments to another party. The context here is that one party is the corporate sponsor with a commitment to pay an agreed schedule of pension contributions to a pension scheme dependent on it. The funds from which the corporate sponsor can meet its annual pension contribution commitments without seeking access to additional external funding can come from two principal sources: (i) corporate cash flow after meeting other commitments; and (ii) liquidity available to the corporate sponsor in terms of surplus cash resources. Corporate net cash flow plus existing surplus liquidity perhaps best captures the concept of the corporate sponsor s ability to pay pension 5 P a g e

6 contributions in any given year without further external funding. External funding may be available in some circumstances, but probably not where affordability of pension commitments is already in question. A financially correct way to measure or quantify affordability in any given year is to compare two quantities: 1. The employer s resources available to pay pension contributions in the given year, equal to Net Cash Flow plus Liquidity (denoted by A); and 2. The required pension contribution in the given year (denoted by B). If A exceeds B then no affordability constraint on the payment of pension contributions will occur, and there is no consequent impact on covenant risk. If B exceeds A, the corporate sponsor can t afford to pay all of its funding commitment to the pension scheme. In this instance affordability will be constrained by the corporate sponsor s ability to pay. This means affordability may have an impact on covenant risk but it is important to understand the quantitative mechanism by which an affordability constraint would increase covenant risk for pension schemes. The impact of constrained affordability can be assumed to typically be a re-negotiation of the contributions schedule and a reduction in pension contributions to a level which the corporate sponsor is able to pay. Unlike a loan or bond, the pension creditor is an open account with considerable repayment flexibility, within regulatory constraints, and it is this which underlines the importance of incorporating affordability into a covenant risk quantification methodology. A reduced pension contribution schedule will push out in time the point (tz) at which scheme funding self-sufficiency is expected to be achieved based on the actuarial model of the scheme. This shifts tz rightwards on the default curve exposing the pension scheme to a longer default curve and additional default risk to which it would not be exposed if pension contributions were affordable. In addition, a lower level of contributions will be very likely to increase LGD. The impact of constrained affordability on covenant risk can therefore be measured and quantified using debt risk measurement techniques adapted for differences in the structure of the pension credit compared to loans or bonds. Affordability: ability or willingness to pay It may in certain instances be important to distinguish between the corporate sponsor s ability to pay and its willingness to pay. The concept of willingness to pay reflects the relative priority determined by corporate boards between discretionary uses of cash flow and liquidity. For the purposes of this paper and quantifying covenant risk it is ability to pay which is the determining factor and delivers the affordability constraint which impacts covenant risk. Unwillingness of a corporate sponsor to pay pension contributions may have important consequences for pension schemes but is principally a matter of commercial negotiation between the corporate sponsor and the pension scheme. Quantifying covenant risk attributable to constrained affordability The impact of constrained affordability on the point in time when funding self-sufficiency is reached will need to be expressed as the number of additional time intervals or years spent on the default to reach the point of funding self-sufficiency. To obtain this number would require inputting into the actuarial model of the scheme a revised pension contribution schedule which reflects the risk exposure to the scheme from constrained affordability. A risk-adjusted pension contribution schedule could potentially be derived from estimating the frequency of an affordability constraint arising and multiplying by the anticipated impact of the affordability constraint on pension contributions. For the sake of argument, this would be expressed as a percentage reduction in contributions compared to the agreed contribution schedule. This methodologically-consistent approach to risk quantification then presents an issue because there are 6 P a g e

7 no existing datasets from which (i) the frequency of occurrence of pension affordability constraints ( PAC ) can be derived, or (ii) the expected impact on contribution schedules ( IOC ) in terms of a percentage reduction in pension contribution rates, can be derived. The two standard approaches to resolving this type of issue in the absence of an evidence-based statistical approach are (a) to review adjacent datasets and then map over the relationship observed for these datasets (pending building a more relevant dataset over time which would be directly evidence based) and/or (b) building a predictive model based on tried and tested probability modelling techniques (which again can be statistically tested for predictive ability once a relevant dataset has been collated). The PAC does have an existing adjacent dataset in the form of loan delinquency data. In debt and credit markets delinquency occurs when loan repayments are delayed. One could therefore borrow delinquency data for the relevant category of corporate credit from external rating agency datasets to use as a proxy for PAC. Alternatively and perhaps more simply one can draw on the standard approach to analysing the relationship between delinquency and default in debt and credit markets and assume a stable relationship over longer time horizons between delinquency rates and default rates. For example, anecdotal evidence from Gazelle s recent study 2 of the past 25 years covenant experience of the FTSE 100 constituents suggests an approximate ratio of 3:1 of identified corporate financial stress to default. Establishing the accuracy of such a multiplier and the stability of relationship to default frequencies represents work that would need to be done over many years to develop relevant datasets for pension creditors. The lack of datasets for PAC or IOC thus leads resolution in the direction of probabilistic modelling. Probabilistic scheme funding outcomes In the absence of historical or statistical evidence for the frequency and impact of affordability constraints, a probabilistic approach is required which is able to compare the uncertain financial resources (A) of the corporate sponsor to prescribed pension contributions (B) for each year or time interval (t) to the point of scheme self-sufficiency (tz) and thereby to quantify the expected frequency and impact of affordability constraints on the payment of pension contributions. The uncertain variable is employer financial resources A. There is opportunity here to select a range of dynamic models for employer financial resources from the simple to the complex. The MT model selects corporate annual net cash flow as the key random variable. Covenant risk value is derived from a stochastic model This paper has identified the three principal components needed to work together to quantify covenant risk: the ratings approach to quantifying corporate default risk (incorporating a default probability dataset and insolvency analysis for the pension creditor); the specific scheme actuarial model defining the term of default exposure in terms of funding self-sufficiency; and the uncertain ability of the employer to pay pension contributions in full, which in turn impacts funding selfsufficiency. Funding Outcomes t = j Employer Default Part Payment Full Payment t = j Accessible at 7 P a g e

8 The quantitative integration into a single financial model of these three elements, as in MT, enables stochastic examination of the probability and impact of three scheme funding outcomes for each year or other time interval: default, part-payment of pension contribution (resulting from an employer affordability constraint), and payment of pension contribution in full. This threefold scheme outcome is displayed in the diagram above. A large number of simulations of the scheme are run under which some simulations successfully reach self-sufficiency with no employer default, and others default and the LGD is recorded. Progress towards funding self-sufficiency is exposed to the impact of affordability constraints if A is less than B. The Covenant Risk Value is the average loss over all these schemes. It is equivalent to the fair economic price one would pay an insurer to take the covenant risk away. Towards an integrated approach comprising covenant, investment and liability risks The principal purpose of this paper is to demonstrate that a consistent and objective methodology for quantifying covenant risk can be achieved by adapting default risk quantification techniques for use by pension creditors. Because the methodology integrates the dynamic scheme actuarial funding model with a dynamic model of employer resources it provides covenant risk with a link to investment performance and scheme liabilities. The quantitative impact on covenant risk value of changing investment performance assumptions or liability cash flows can now be easily examined with MT. If investment risk is expressed using mean and volatility (i.e. as a distribution of investment performance outcomes) there are perhaps two ways to approach the quantitative relationship with covenant risk. One is to calculate the covenant risk values across the variance of investment return outcomes, while the other is to allow covenant risk and investment risk to interact by modelling them together as two uncertain variables. Potential integration with Asset Liability Modelling The stochastic modelling of uncertain pension contributions can now in principle be integrated with stochastic modelling of uncertain investment performance. Current investment consulting practice is to run stochastic models of uncertain investment performance (ALM) which incorporate deterministic assumptions about contributions. Combining existing ALM models with MT or equivalents should give a more realistic and prudent (but probably wider) variance of outcomes. This should prove very important in setting the correct scheme funding targets which is an area currently attracting considerable focus. Implications for the UK pension regulatory framework The quantitative approach and methodology developed in this paper offers a consistent and objective framework through which to develop professional practice in the direction of introducing quantitative risk measurement and management methodologies to match for pension credits what is already the norm in debt and bond markets and their supervision. The quantitative approach outlined appears to coincide very closely with the guidance issued by tpr for covenant assessment in terms of affordability analysis, insolvency analysis and linkage to investment performance. Importantly it now provides a quantitative framework which encapsulates and integrates these elements. In theory the PPF holds a portfolio of pension credit exposures similar to a bank loan portfolio and the introduction of covenant risk quantification by schemes if reported to PPF could enhance the efficiency with which PPF is able to manage its risk exposure. 8 P a g e

9 Gazelle Corporate Finance is a leading independent covenant assessment and advisory firm focusing on larger schemes and complex situations. Cross-practice discussions within the firm identified the idea that a quantitative methodology might be developed incorporating credit risk exposure which could then integrate the key areas of qualitative covenant assessment into a single quantitative model. The author of this paper Simon Willes developed the MT (Mousetrap) model with assistance from Alex Barrell of Trinity College, Cambridge and Matt Hitchings, and with input and review from Paul Thornton OBE and Donald Fleming. MT is a commercial financial model owned by Gazelle MT LLP and is available to other professional firms under license (subject to commercial agreement) as a cloud application. For further information please contact: Simon Willes simon.willes@gazellegroup.co.uk 9 P a g e