The FCA’s Thematic Review into Retirement Income highlights some important points – and some key dangers advisers need to be aware of – in using cashflow modelling in retirement-income advice.
Cashflow modelling is commonly used to test the suitability of retirement income strategies. The assumptions within these models vary. This is not in itself a problem (and indeed, not varying them when client circumstances suggest they should be varied is poor practice).
However, the Review notes a major problem: a lack of a robust methodology for determining what rates to apply, when to apply them, and how to demonstrate to the investor that it is a variable, not a judgement on their likelihood of the strategy succeeding.
We wrote about this, and other dangers of complex, over-engineered, stochastic cashflow models in 2021. Given how the Thematic Review has brought these issues back into the spotlight, we’re republishing it below.
Cashflow modelling and the costs of complexity
Cashflow modelling has long been at the core of financial planning. It is where diagnosis of the past and predictions of the future come together to prescribe actions for the present. In effect, it is an attempt to solve a puzzle – how should you account for long-term uncertainty in your short-term actions?
A scattering of puzzle-solving approaches has evolved, though each boils down to a cost-benefit analysis of a lifetime of ins and outs: of income, inheritances, goals, and gifts; and appropriately weighting the likelihood and influence of each.
Evolution repurposes existing features for new functions, however. The further a model evolves, the more important it becomes to check how well it is answering the questions it set out to. Increased complexity is a cost that the new benefits need to justify. What are the costs and benefits of your cost-benefit analysis? Is the model still a means, or has it accidentally evolved into an end in itself? Is your solution elegant, or overengineered?
With its increasing focus on stochastic approaches, cashflow modelling has evolved into an imitation of something useful. Faced with the impossibility of predicting the future, stochastic tools instead settle for predicting all possible futures as a proxy. This can be useful for testing sensitivity to an isolated variable but, as a planning tool for a real investor and a real set of investments, it can prioritise complexity over an answer, and spurious theoretical precision over practicable real-life outcomes.
A cashflow process is not an end in itself. Judged against what it is trying to achieve, complex stochastic tools are less use than you may think for a number of reasons.
They do not tell you what you need to do
Any approach must answer the question, what is the right level of risk to take right now? Suitability along an investment journey is essentially asking repeatedly, as goals and circumstances change: to what extent should risk levels be adjusted?
The near-term future is not a stochastic model's comfort zone because the ideal response to each possible future is different. The reliability of decades of data dwindles in individual cases. Betting on one person's experience of the next five years behaving like the average of the past 100 is playing with high stakes and long odds.
It is far better to have a reliable answer for a smaller set of possible circumstances, and to focus on adaptability, to enable changing course quickly when possible futures become observable realities. On changing terrain, simple and flexible is safer than complex and brittle.
The right level of risk to take right now is a question for risk capacity, not a crystal ball. What is the total risk suitable for the investor, what risk are they already exposed to with non-investible assets, and what is left over for investible assets, accounting for the likelihood of future changes?
As an example, someone may have a large – perhaps unaffordable on current projections – spending goal 10 years in the future. The goal does not influence risk capacity much since, at 10 years away, the right thing to do is to stay invested regardless. If this goal were only four years away, however, the right thing to do would be to reduce risk (if high priority), or delay, reduce, or eliminate the goal (if it is relatively unimportant).
They do not tell you what you really want to know
An investor needs to know whether a given system of long-term goals is affordable – in other words, will they run out of money? Complex calculations are fragile because long-range forecasts are heavily sensitive to a myriad of opaque assumptions about the behaviour of both markets and investors. Even when investment variables are allowed to float in stochastic models, the other half of the inputs – investor variables – are assumed to be implausibly static.
The further you move from simplicity, the more you risk sacrificing the reality of robustness for the illusion of accuracy. Sufficiently clear answers to what we really want to know – when might I run out of money? – do not need stochastic projections. They can be calculated with simple tests under a few understandable scenarios. Such tests can enhance understanding without inspiring overconfidence, and indeed the sophistication of stochastic tools in this instance tends to add confusion rather than clarity.
The confidence complex models inspire can be misleading
The ultimate aim of models is to inspire client comfort and confidence with investing. Ideally, reassurance is derived from easy-to-understand answers to common 'what if?' questions and the process behind them. Complex solutions, however, can rely instead on overfitting and overconfidence – especially where a model looks smarter than your average spreadsheet.
Tackling the uncertainty of investment inputs with thousands of projections can lead to unwarranted confidence in outputs – results and plans that are spuriously precise, and that discourage the sort of deeper examination of lifestyles that makes the plans so valuable in the first place.
There is an underlying belief that more moving parts equals more accuracy, and stronger 'evidence' of meeting regulatory requirements. This is often based less on the sort of client understanding the regulations are designed to inspire, however, and more on the impressive-looking spelling of 'stochastic'.
Simple does not need to mean basic
Complex stochastic approaches can add value, but that value is limited, requires sophisticated users and involves a disproportionate cost of time and effort relative to the benefits provided. Even if a stochastic tool is good, the more complex it is, the more likely it is to be used badly. And for all its sophisticated appearance, complexity often does not provide much additional actionable information – it neither tells you better what to do, nor what you really want to know.
A simple model that supplements and supports, rather than dictates the terms of, the adviser-client relationship, and that works dynamically to encourage flexible responses to events, not complex predictions of them, is the real sophisticated solution to the cashflow puzzle.
Thumbnail picture by Maxim Hopman on Unsplash
