The design of retirement products is an increasingly pressing challenge for many pension plans.
In Australia, policy is being formulated that will require superannuation funds to offer a Comprehensive Income Product for Retirement (potentially to be named ‘MyRetirement’ products) to members at the point of retirement.
Progress has been mixed. A survey taken at a [i3] Pension Portfolios Forum in late-October 2019 indicates that only a small minority have developed these products, while the majority are considering how to proceed (see below).
While there are various reasons for the limited progress, one factor is the technical difficulty of designing retirement products. A range of hurdles exist.
First, retirement is complex. In moving from accumulation to decumulation, the problem shifts from building wealth to drawing down on assets over time with uncertainty over time of death and spending needs.
This requires evaluating random outcomes over multiple periods while addressing the drawdown as well as investment strategy, both of which should ideally be dynamic in nature.
Second, member heterogeneity is manifest. Members can differ along dimensions such as their required income, account balance, assets outside of the retirement account, access to any pension, household status … the list goes on.
This means that one-size-fits-all products are tenuous, and the need for a range of product options that cover a variety of member types.
The third hurdle is communicating the available options to members in a way that they can understand.
In a recent paper, I suggested an approach to designing menus of retirement products that addresses these hurdles. Two main concepts sit at the foundation of the approach.
First is the use of ‘tailored’ utility functions to characterise member preferences over outcomes.
The second is framing around member attributes such as their outcomes of concern, assumed tolerance for poor outcomes, balance, other available assets, personal status, etc. These attributes provide the basis for member segmentation, utility function specification and member communication, with the latter supported by descriptions of product features and potential outcomes.
The approach entails four stages:
- Segment into member types – Member types for which products will be designed are identified and characterised by selected attributes.
- Specify a utility function for each member type – The utility function provides a system for placing ‘scores’ on potential outcomes so that products can be evaluated.
- Model the underlying investment and drawdown strategy – The modelling involves simulating potential outcomes that arise from adopting a particular investment and drawdown strategy. The total score arising from applying the utility function supports either identifying the ‘optimal’ strategy, or comparing selected candidate strategies. Meanwhile, familiar metrics such as the median outcome, shortfall measures and failure rates are generated as a check, and as a way to describe the outcomes that the product generates.
- Communication to members – This is based around: using attributes to convey the type of member for which a particular product is designed; setting out the product features including the underlying asset mix and drawdown strategy; and, describing the outcomes that might be expected such as expected income, how long that income might be supported, and the downside if investment returns are poor. The idea is to recommend a product to a retiring member based on what is known about that member; while making it clear that other options are available in case they happen to be a different type to what is assumed.
Further details can be found in my paper, which illustrates how the approach might work. It does this by proposing six hypothetical member types, and going through a simplified example of the modelling and communication process for a selected member type and resulting product.
In the remainder of this article, I will explain how utility functions can form a key part of the toolkit for retirement product design.
Utility functions offer an effective way to compare strategies and hence products when the outcomes are too complex to be readily addressed through simple metrics.
Retirement outcomes typically accrue over multiple periods and may span multiple objectives, e.g. generating an income stream plus a bequest. A utility-based approach can support condensing all possible outcomes across and through time into a single score by aggregating utility from the stream of income and any bequest, placing a discount on the latter.
Generating a single score that embeds the trade-off between better and worse outcomes avoids having to make the trade-off by either considering a matrix of metrics, or cherry picking a few measures such as the average outcome versus a shortfall measure that is only a snapshot of the entire distribution of outcomes.
A second key advantage is the potential to tailor to different member types through assigning different utility functions. You start with the member type, identify what outcomes they are likely to be concerned about, and then select the utility function accordingly.
For instance, a reference dependent utility function may be used where there is an income target, and another utility function for balance-focused members. This flexibility can support designing menus of products that cater for members with various objectives and preferences.
I am not suggesting that the more familiar metrics should discarded. Utility functions provide a better engine for analysis and product selection, while metrics provide the dashboard that reveals what the product might deliver. Both should be used in tandem.
Either way, assumptions must be made about what matters to the member, and their tolerance for poor outcomes. What I am arguing is that embedding the trade-off in a single score arising from a utility function is more effective for selecting retirement strategies.
Designing retirement products is not easy, and requires advanced technical methods to deal with the complex, multi-period nature of the problem. I hope that the approach summarised here points to a way forward.
The paper can be located on SSRN at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3504313
Geoff Warren is Associated Professor at the Australian National University.