Why Lifecycle Investing Is an Analytics Problem, Not Just an Investment Problem

For CPF providers, building the glidepath is only the beginning. Measuring whether it will work, across 30 to 40 years of economic scenarios, is the harder challenge.

Most commentary on Singapore's CPF Lifecycle Investment Scheme focuses on the glidepath: how the asset allocation should shift from equities to bonds as members approach retirement, what the target dates should be, and how aggressively to de-risk. But there is a harder question sitting beneath them, and it is the one that independent investment consultants will spend most of their time on.

How do you know, with any rigour, whether your glidepath will actually work?

This question matters more now than it did before the CPF Board's announcement confirmed that selected providers will be evaluated by independent investment consultants. Those consultants are not going to look at a glidepath illustration and approve it on aesthetics. They are going to ask what methodology produced it, what stress tests were applied, and what the analytics say about outcomes across a realistic range of economic scenarios. Providers who have not thought carefully about the analytics infrastructure underneath their investment design will find that gap very difficult to close under pressure.

Our previous piece in this series covered what selected providers need to be doing between now and early 2027. This post is for CTOs, heads of product, and digital transformation leads who need to understand what kind of analytical infrastructure supports credible lifecycle fund design, and why the tools most institutions reach for first are the wrong ones.

Why are standard optimisation tools the wrong starting point

The instinct at most institutions is to begin with mean-variance optimisation, the Markowitz framework that has been taught in finance courses for decades. It is familiar, it is well-documented, and it produces clear portfolio allocations. For a lifecycle fund, it is also structurally mismatched to the problem.

Mean-variance optimisation treats gains and losses symmetrically. If the expected return of a portfolio rises by 2%, it treats that exactly as desirable as a 2% reduction in volatility. For a member who is saving for a defined retirement sum, this symmetry is not how outcomes actually feel. The member who ends up SGD 15,000 short of the Full Retirement Sum (FRS) has not had a symmetrically bad experience to the member who ends up SGD 15,000 above it. The shortfall is the problem. Variance, as a measure of risk, cannot see this.

Expected Shortfall (ES), which is sometimes called Conditional Value at Risk, or CVaR, is a more appropriate measure for this context. Rather than treating all deviations from the mean equally, Expected Shortfall focuses on the average outcome in the worst 5% of scenarios. For a provider designing a product whose purpose is helping members reach the FRS, understanding what happens in those adverse cases is far more useful than minimising variance across all cases. This is what regulators and sophisticated investors now typically expect to see in retirement product design.

Why deterministic glidepaths miss the point

A deterministic glidepath is easy to present. You show a chart of equity allocation declining smoothly from 85% at age 25 to 10% at age 65, and it tells a clear story. What it can’t tell you is whether that allocation pattern will produce a sufficient retirement balance across the distribution of market outcomes a member might actually experience.

The reason is sequence-of-returns risk. A member who experiences a significant market drawdown in their early fifties, a decade before retirement, when the portfolio is still heavily invested in growth assets, faces a fundamentally different outcome than a member with identical average returns but no drawdown near retirement. The timing of losses, not just their magnitude, determines whether the glidepath delivers. A deterministic model cannot capture this because it does not model the path, only the endpoint.

What is required instead is a stochastic simulation engine, one that generates thousands of distinct economic scenarios, each representing a plausible path for equity returns, bond yields, and inflation over a 30 to 40 year horizon. The glidepath is then run through every one of those scenarios, and the result is a probability distribution of outcomes rather than a single projected number. That distribution is what allows a provider to say, with statistical rigour, what proportion of members are likely to reach the FRS, what their expected shortfall looks like in adverse cases, and how the glidepath performs under conditions that look like past crises.

What the model needs to get right about markets

Not all stochastic simulation engines are equal. The quality of the output depends entirely on whether the underlying Economic Scenario Generator (ESG) accurately models the statistical properties of real markets, and this is where many generic tools fall short.

Real markets are not normally distributed. They exhibit fat tails and volatility clustering: extreme events that occur more frequently than a Gaussian model would predict, and periods of high volatility that tend to follow each other rather than appearing randomly. Correlations between asset classes shift under stress: assets that appear uncorrelated during calm periods can fall simultaneously during a market crisis. A model that assumes normal distributions and stable correlations will systematically understate the risk in the lower tail, exactly where a retirement product needs its numbers to be most reliable.

For CPF specifically, the ESG must also reflect Singapore's asset universe: the STI, global equity indices in SGD terms, Singapore government bonds, and the interest rate environment for CPF accounts themselves. A model calibrated to US or European market conditions will produce scenario distributions that are incorrect for this product category.

The involvement of independent investment consultants in the evaluation process has a specific implication that providers should think through carefully. Although the consultants will be reviewing the outputs, they will also ask how they were produced.

That means the analytics infrastructure needs to support full methodological transparency. The ESG calibration process: how the model was fitted to historical data, how tail probabilities were estimated, which assumptions were made about future return distributions, these all need to be documented and defensible. Stress-test outputs need to show performance under named crisis scenarios, not just abstract percentile bands. FRS attainment probabilities need to distinguish between the eligible balance that actually counts at age 55 and total modelled wealth. These are are substantive tests of whether the analytical framework is fit for purpose, not just presentation choices.

An audit trail also matters. Every projection that is shown to a member or submitted as part of governance documentation needs to be reproducible, the same inputs should produce the same outputs on demand. This requirement effectively rules out spreadsheet-based approaches which cannot guarantee reproducibility at scale.

The infrastructure question providers cannot defer

Building this infrastructure from scratch is a multi-year undertaking for most organisations. A properly calibrated ESG for Singapore's asset universe, integrated with Monte Carlo simulation at the scale required for real-time member projections, the kind that updates instantly as contribution patterns or market conditions change, sits at the edge of what internal quantitative teams can realistically deliver within the timeline available.

The practical question is not whether to have this infrastructure, but how to get it. At Kidbrooke, the KidbrookeONE platform provides exactly this capability through a stateless, API-first architecture: an Economic Scenario Generator calibrated to reproduce fat tails and volatility clustering, Monte Carlo simulation running 5,000 scenarios across 720 time steps per recalculation, Expected Shortfall and FRS attainment probability outputs, and full audit trails for governance. The platform stores no end-customer data, which also matters for a scheme handling retirement savings at national scale.

The deployment timeline is realistic. Our work with Skandia, one of the Nordics' largest pension providers, managing SEK 863 billion in assets, took their platform from concept to a live digital advisory service in four months. That kind of speed is available because the analytical infrastructure already exists; providers are integrating a validated engine rather than building one.

The glidepath is the product. The analytics is what makes the product credible.

This is the point that providers who frame their CPF submission primarily as an investment strategy question may be missing. A well-designed glidepath is necessary but it’s not sufficient. The consultants evaluating submissions know what a stochastic simulation should look like, they know the difference between normally distributed models and fat-tailed ones, and they will spot the gap between a glidepath illustration and a tested analytical framework.

Providers who have built, or integrated, the infrastructure to produce probability-based outcomes, Expected Shortfall figures, and full audit trails are not just better prepared for regulatory scrutiny. They are better positioned to run the scheme effectively once selected, to demonstrate to members that their retirement savings are being managed rigorously, and to adapt as Singapore's market conditions evolve over the decades the scheme will run.

If you are currently working through what your analytics stack needs to look like to get there, get in touch.

Kidbrooke is a financial technology company providing unified investment and wealth analytics through KidbrookeONE, an API-first platform serving pension providers, asset managers, and wealth platforms.