While pensions management strategy becomes increasingly sophisticated and high-profile, the quality of the foundation of these decisions – the pension management information – can vary widely. Neil Wharmby explains what to look for when assessing what models to use when developing and implementing your company’s pensions strategy.
The management of defined benefit schemes has never been more complex or as important to a sponsor company’s finances, as it is today.
In particular, the low-yield environment has increased the relative size of liabilities compared to the assets for schemes which are not yet fully hedged. Many companies and trustees are increasing the sophistication of the models used to develop and implement their pension management strategies. But beware, the wrong choice of pension model may weaken the effectiveness of your pension strategy!
The models employed embed subjectivity in both their approach (methodologies) and their inputs (assumptions and data) that can skew the outcome of your chosen strategy. This can result in inefficient use of scarce resources such as cash and management time.
If we assume that the model being considered is technically robust (the model can do what it claims to do) then how do you determine which model is best for your needs?
Our four top tips to help you assess the suitability of pension scheme models can be found below:
1. Data – are you trying to crack a nut with a sledgehammer?
A lot has been said in pension circles about the benefits of an on-demand ‘full valuation’ versus what are commonly referred to as ‘roll-forward models’. The distinction is between the data used: in the roll-forward approach, the starting point is the data from the previous actuarial valuation, often (but not always) adjusted to allow for material changes, while the full valuation requires a fresh extract of the underlying pensions data ‘from scratch’. The on-demand full valuation may seem like the better option, but is it really?
We have found that instead of increasing accuracy, new data extracts can sometimes lead to greater error than the experience they otherwise reveal.
Extracts, particularly for large schemes are not always robust and time is needed to ensure that the data is complete and fit for purpose.
But, the real issues arise when assessing how much experience this new data brings, and how many of your previous assumptions it replaces? Let’s look at emerging inflation experience as an example. Does the data reflect the last annual increase, or have allowances been made for subsequent known inflation changes? Has a salary increase date passed, and how much of this increase was already incorporated into the previous results through already known market inflation data and assumptions? Is the inflation built into the liabilities consistent with the inflation information inherent in the asset pricing? Failure to adjust for points like these could easily have left you looking at a funding position with an error of 5% or more over the last few years.
In contrast, the impact of, say, basic membership experience – retirements, deaths, commutation profits – the very sort of experience that the new extract is aiming to capture, might only be worth around 1% or so over a three-year period for many stable schemes.
Taking all of this into account, rolling forward robust results might actually be more accurate than the potential errors of mixing new or incomplete data with past assumptions and models.
2. Dynamic assumption modelling is now a ‘must-have’
A dynamic approach to modelling assumptions is now a must-have. By this we mean that you should be looking for a model that can accommodate term-dependent assumptions, has the ability to track these over time in conjunction with changing yield curves and can provide term-dependent asset return expectations.
Many schemes have already moved to valuing their liabilities in this way in order to better align with investment strategy decisions and more transparently assess the effectiveness of both liability hedging and return-seeking strategies. For those that haven’t yet moved, there is a clear trend in this direction as schemes mature close to future accrual, or adopt more closely-hedged investment strategies.
If you’re thinking – “I use a yield-curve approach already, this doesn’t apply to me” – then we would challenge you to consider whether the modelling of that approach over time is truly dynamic. The common shortcut of modelling using ‘mean term’ adjustments can lead to errors in excess of tolerances for many purposes when yields move materially, as they have over the last 18 months. In addition, if your modelling assumes constant margins over gilt yields then are you confident that this a reasonable reflection of changing corporate spread and movements in equity pricing?
On a more detailed note, a truly dynamic model should be able to automatically adjust for caps and floors as the inflation curve changes, allowing you to model the increases in assumptions that are used to measure your liabilities consistently with any Limited Price Indexation assets held (or to provide more transparency about the effect of holding such assets).
The use of dynamic assumption models automates the assumption-setting process as well as reduces the cost of configuration and recalibration. This approach also provides the ability to track objectives over a longer period by making a clearer distinction between market changes and assumption changes which are made as a result of changes in actuarial approach. There is no longer the need to worry about the darkness of the actuary’s ‘black box’.
If you are using the results from these models to set your strategy or implement key changes then it should go without saying that the approach used to derive those results should be transparent and accessible.
If you’ve read the preceding paragraphs and realised that you’re not sure how the assumptions in your models are set, then now might be a good time to ask a few questions. In particular, are the assumptions and modelling parameters clear and appropriate for the purpose for which they are being used? Indeed, do you know what purpose the figures and assumptions have been designed for?
Actuaries required to comply with technical actuarial standards issued by the Financial Reporting Council (FRC) must satisfy themselves that the models they are using to inform their decisions are fit for purpose and appropriately documented. Would the key decision-makers on your board expect any less of you? (This latter point is particularly relevant given the application of the FRC’s new technical actuarial standards to all technical actuarial work from 1 July 2017. We expect that all companies would seek consistency of compliance with FRC standards in all aspects of their financial governance – pensions disclosures included, whether or not prepared by an actuary.
4. Value for money
Full valuation and asset liability modelling systems certainly have their place but they come with a reasonable price tag. It is important to not only consider the cost in terms of licensing fees, but to also look at the resource needed to construct and maintain the models and assumptions, as well as to provide appropriate professional review of any figures used. So how can you reduce your costs of pension scheme monitoring?
The first stage is to identify the objectives of your modelling and consider what the key drivers are for achieving those objectives. For example, do you really need the risks and costs associated with regular data extracts if you are only looking at compliance reporting? Would incomplete or inaccurate asset feeds lead to misleading results and potentially damage investment decision making? Do you really need to know how many retirements have taken place if you are making decisions around asset allocation? Without this thought process, there is a danger of purchasing a model that is either very expensive, not fit for purpose or, in some cases, both!
So what model should I use?
Easy. One that‘s fit for purpose and affordable. Pension schemes are complex, but that doesn’t mean that your life needs to be too! Well-constructed models are often more accurate than badly parameterised or unchecked full valuation systems. Which would you rather?