The current level of underfunding of corporate defined benefit (DB) pension plans has raised awareness of the risks inherent in the plans – risk to plan members in the case of insolvency, risk to plan sponsors relative to the need to make large contributions, risk to governments (or their agencies) should they have to bail out the plans, and risk to shareholders should firms whose assets they hold have to make large contributions.
In the US alone, underfunding at corporate DB plans is estimated to be in the range of $450bn (e346bn). The Pension Benefit Guaranty Corporation, a self-financing federal agency that insures corporate pension plans, was running a deficit of $23bn at fiscal year-end 2004 with an additional “reasonably possible” exposure of $96bn. By some estimates, the underfunding at S&P 500 companies represents 40% of the forecast profits for firms in the popular index for the year 2004. At question is the management of some $1.5trn in DB pension fund assets in the US alone.
The inherent risk in DB pension plans suggests the need for a wider adoption of statistical techniques and computer models to measure and help control risk. It was the increased awareness of risk after the 1987 crash of the stock markets that was behind the adoption of computer-based risk management systems in the banking sector. A recent study by Yale University professor Frank J Fabozzi and The Intertek Group looks at how pension funds (and regulators) in four countries where corporate plans play a large role in pensions provision , namely the Netherlands, Switzerland, the UK and the US, are responding to the challenge of managing risk.
The report, ‘Can Modelling Help Deal with the Pension Funding
Crisis?*, is based on conversations with 28 funds responsible for managing a total of e334bn in assets; average assets under management by participating funds ranges from e7.3bn in Switzerland to e15bn in the Netherlands.
A primary objective of the study was to determine if there has been a shift away from a focus on managing for returns to liability benchmarking. Liability benchmarking requires searching for correlations between assets and liabilities, so it calls for the modelling of both. The study found that a shift has indeed occurred in the Netherlands, where almost three quarters of the participants either have already adopted liability benchmarking or are considering ways to incorporate liabilities into their benchmarks.
This is a reflection of a new regulatory framework which introduces market valuation of liabilities and assets and mandates their integration. Participating funds in the UK and the US reported that the accent remains on long-term investment returns. Most of these funds - with roughly 70% of their total assets in equities - cited strong cash flows and/or high funding ratios as the motivation for not embracing liability management.
Another important objective of the study was to understand to what extent full-fledged ALM studies are done. We consider as full-fledged an ALM study based on computer models that project future liabilities and returns, integrate their ‘paths’, explore alternative scenarios, and suggest risk-return trade-offs. Again the study found significant differences at the national level.
In Switzerland, the UK and the US, ALM studies are typically performed by external consultants every three to five years; about a third of the participants in each country mentioned doing an ALM study in an informal or qualitative way. (It should be noted that Swiss funds continue to be managed conservatively: four of the five Swiss firms participating in the study have less than 30% of their assets in equities.)
Full-fledged ALM studies was the rule among the Dutch funds, with a trend towards yearly ALM studies. A widely appreciated benefit of a full-fledged modelling exercise is that it fosters a good process: by delivering quantitative-based information, it allows fiduciaries to express their risk tolerance and understand what might happen on the downside. The impact of various investment, contribution, indexation and financing policies can be thoroughly explored.
Funds were asked how they
evaluated today’s modelling tools. For the most part, the modelling of liabilities is considered to be the domain of external consultants. There is no ‘picking apart’ of the models as was recently the case with market and credit risk models in the banking sector. However, five of the participants (three of which are industry-wide Dutch funds) mentioned that they now run their own liability studies in-house, using commercially available or proprietary software. As the Netherlands moves towards market valuations for liabilities, Dutch funds are also building up experience in modelling interest rates: roughly half the Dutch funds participating in the study have done work on this internally.
Returns modelling is widely done in the US and in the Netherlands; in Switzerland and the UK, qualitative forecasts are more typical among our sources. In the US, where the focus remains on investment returns, seven of the 10 participating firms model assets in-house with third-party software, using a top-down and/or bottom-up approach. The typical time horizon is five years out. Among large Dutch firms, returns are typically modelled at multiple time horizons. The new regulatory framework, which requires the explicit indication of assumptions on investment returns, will reinforce this as solvency will be tested at one- and 15-year time horizons.
The technique most widely used to evaluate the impact of possible future movements is scenario generation, which is used by all the Dutch funds and a great majority of US funds. Because scenario generation allows for what-if reasoning, it is considered an easy tool to facilitate discussions with trustees. Though many funds mentioned preferring simulation to optimisation, optimisation is performed (typically by external consultants) at two-thirds of the participating funds. Some funds mentioned combining optimisation with techniques to capture tail events or unstable correlations.
Risk management is gaining increasing attention. Among those questioned in the report who do no risk modelling in the ALM process (40%), a number of funds mentioned that risk is now beginning to receive more attention. Scenario generation is widely used in risk management, with Monte Carlo techniques generating the scenarios. The new regulatory framework which comes into effect in the Netherlands in 2006 will require that funds stress test their funding levels on multiple time horizons using both shock and trend scenarios. Given new asset classes and today’s volatile markets, some participants also mentioned that they are beginning to look more closely at skewness and kurtosis.
One problem widely commented on is the failure to integrate the results of the ALM study in the decision-making process. In many instances, models are only ‘cosmetic’. However, familiarity with modelling is growing: just under half of the funds participating in this study mentioned having hands-on experience with models. The need to build up in-house expertise on modelling issues and to integrate model results in the decision-making process has been recognised, particularly in the Netherlands.
Pension fund management is subject to many sources of uncertainty: demographic trends, inflation, interest rates, markets, etc. In other words, pensions are subject to risk. Using theory and statistical analysis, modelling reduces risk. Pension funds can be managed conservatively and safely without modelling, but as funds seek high returns investing in equities, hedge funds and other risky asset classes, modelling becomes increasingly important in order to deliver a better process with risk under control.
Caroline Jonas is research director at The Intertek Group in Paris
*The report ‘Can Modelling Help Deal with the Pension Funding Crisis?’ is available at www.theintertekgroup.com