Methodology

Most retirement calculators use one of two approaches: a deterministic projection (assume 6% return every year) or a simple average (average historical return applied uniformly). Both approaches are dangerously misleading.

The problem is sequence of returns. A retiree who experiences a severe market crash in their first three years of retirement is in a fundamentally different position to one who experiences the same crash in year fifteen\u00a0\u2014 even if the average annual return over the whole period is identical.

Consider two retirees with identical portfolios and withdrawal rates, one retiring in 1999 (dot-com crash immediately) and one retiring in 2003 (crash already happened). The same average return over 30 years produces radically different outcomes because of when the bad years fell.

Monte Carlo simulation addresses this by running thousands of simulations, each using a different sequence of returns. Some sequences are benign. Some include severe crashes at the worst possible time. The survival probability\u00a0\u2014 the percentage of simulations where your portfolio lasts to your target age\u00a0\u2014 is a genuinely informative metric in a way that a single projected figure is not.

Selvox uses three primary data sources:

We do not sample individual years independently. Independent sampling ignores the short-term autocorrelation in returns\u00a0\u2014 the fact that good years tend to cluster with good years and bad years with bad years (momentum and mean reversion effects).

We use block bootstrapping:

1. 1.Divide the historical return record into overlapping blocks of five years.
2. 2.For each of the 10,000 simulations, randomly sample blocks with replacement until the required simulation length is covered.
3. 3.This preserves the short-term clustering of good and bad years while allowing sequences longer than the historical record.

Block size of five years was chosen to capture medium-term economic cycles while providing sufficient sampling diversity across 10,000 runs. The block size is configurable in the advanced assumptions (Step\u00a06 of the calculator).

We also implement contiguous window sampling for the historical stress test\u00a0\u2014 running the plan against every actual 30-year sequence in the historical record to show how it would have performed in each real historical period.

The UK equity return series from Barclays and JST reflects the performance of UK markets that continued to exist and be investable throughout the period. Markets that failed entirely (e.g. Tsarist Russia, pre-war China) are not represented.

This creates an upward bias in historical return data\u00a0\u2014 we are observing the survivors, not the full distribution of possible outcomes.

We apply a correction of \u22120.5% per year to equity returns, following the methodology of Dimson, Marsh, and Staunton (London Business School) as documented in their Credit Suisse Global Investment Returns Yearbook. This is a conservative adjustment\u00a0\u2014 some researchers argue for a larger correction.

This correction is applied by default and can be disabled in Step\u00a06 of the calculator. We recommend leaving it enabled.

Academic citation: Dimson, E., Marsh, P., and Staunton, M. Credit Suisse Global Investment Returns Yearbook (annual). London Business School.

State Pension entitlement is calculated from qualifying NI years (maximum 35\u00a0years for full new State Pension of \u00a3241.30/week in 2026/27).

We project future State Pension income using the triple lock assumption: the greater of CPI, average earnings growth, or 2.5%. The default assumption is 3.0% per year, which we consider a reasonable long-run estimate of the triple lock outcome. This can be adjusted in Step\u00a06 of the calculator.

State Pension deferral is modelled at 5.8% per year of additional income for each year deferred beyond State Pension age (currently 66).

State Pension age changes are not currently modelled\u00a0\u2014 the calculator uses the current State Pension age of 66 throughout. If you believe your State Pension age will be higher (likely for younger users), adjust your inputs accordingly.

The US \u201c4% rule\u201d originates from the Trinity Study (Cooley, Hubbard, and Walz, 1998), which was based on US market data. It does not directly apply to UK investors:

• UK equity returns have historically been lower and more volatile than US returns over the same period.
• UK inflation has historically been higher.
• UK tax rules are different — particularly the treatment of pension income and the Personal Allowance.
• The US study assumed a 30-year retirement. Many UK planners targeting early retirement need a 40+ year horizon.

Selvox calculates a personalised UK safe withdrawal rate for each user by finding the spending level at which the plan survives 95% of historical simulations. This figure varies by asset allocation, time horizon, and tax situation.

Based on the available UK historical data, the equivalent of the US 4% rule for a UK investor with a balanced portfolio (60/30/10 equities/bonds/cash) and a 30-year retirement is approximately 3.0\u20133.5%. For a 40-year retirement, it is approximately 2.5\u20133.0%.

Academic reference: Pfau, W. (2010). \u201cAn International Perspective on Safe Withdrawal Rates.\u201d Journal of Financial Planning. This study found that US-derived SWRs overstate safe withdrawal rates for most other countries including the UK.

We believe in being honest about what Selvox does not model. Current limitations include:

Future tax changes

All projections use current 2026/27 tax rates throughout the simulation. In reality, tax rates will change. We update the tax engine within 48 hours of every Budget but cannot predict future policy.

State Pension age changes

The calculator uses the current State Pension age of 66. This is likely to rise for younger users. Adjust your inputs to reflect your expected State Pension age.

Defined benefit pension complexity

We model DB pensions as a simple inflation-linked income stream. Complex DB arrangements (with early retirement penalties, commutation factors, bridge payments, or enhanced transfer values) are not fully modelled.

Care costs

We do not model the probability or cost of requiring long-term care. For a 65-year-old, the probability of requiring some form of care is significant and the costs are substantial. This is on the roadmap for a future update.

Behavioural factors

The model assumes you withdraw exactly the planned amount each year and never deviate. Real spending behaviour is more complex. The Guyton-Klinger flexible spending rules are available as an option but are not the default.

Inheritance and estate planning

We do not currently model inheritance tax, gifting, or estate planning. The SIPP outside-estate treatment is noted but not modelled.

We validate Selvox\u2019s outputs against official sources:

Income tax and NI

Validated against HMRC’s PAYE tax calculator (tax.service.gov.uk/estimate-paye-take-home-pay) for multiple income levels and scenarios.

State Pension

Validated against the DWP Check Your State Pension service (gov.uk/check-your-state-pension-forecast) for multiple qualifying year combinations.

Monte Carlo outputs

Cross-referenced against published cFIREsim results for comparable inputs as a sanity check on the simulation framework.

If you find a discrepancy between Selvox\u2019s outputs and an official source, please report it to hello@selvox.co.uk with the specific inputs and expected vs actual output. We treat these reports as high-priority bugs.

Planned improvements in approximate priority order:

• Couples mode — full joint optimisation (near term)
• Defined benefit pension commutation modelling (medium term)
• Care cost probability modelling (medium term)
• Inheritance tax and estate planning module (longer term)
• Variable annuity comparison (longer term)
• CAPE-adjusted return expectations (longer term)

If there is a feature you need that is not on this list, please contact us. User requests directly influence the roadmap.