Assumptions & limits

Every projection of a forty-year future is a model, and every model is a simplification. RetireSafe has to simplify; the useful question is where, and whether those simplifications could mislead you. This page is the honest account: what the model leaves out, where it takes shortcuts, and how to read the results with all of that in mind.

What the model leaves out

• Tax is modelled, but not completely. Your country’s scales, allowances and thresholds are applied to your actual position each year, across twenty countries, and the tax page lists exactly what each one covers. What is left out is consistent across all of them: local rates within a country are averaged rather than asked about, elections a tax return would let you optimise are fixed at the branch that suits a typical retiree, inheritance tax is a separate system we do not touch, and the years before you retire use a simplified cash flow that understates what a contribution to a retirement account is worth. Where a figure matters to a decision, check it with someone who files returns for a living.
• No behavioural feedback beyond the rules you choose. Real retirees adapt: they cut back in a downturn, splurge in a boom, change plans. The model only adapts in the specific ways its withdrawal rule and spending-cut logic prescribe. It does not assume you’ll be cleverer (or more reckless) than that.
• Health is modelled as mortality plus long-term care. There is no separate modelling of disability, chronic illness, or medical events beyond the healthcare cost curve and the long-term-care episode.
• Twenty countries are calibrated, and market assumptions are not among them. Tax rules, life-expectancy tables, care costs and property assumptions all ship per country. Investment returns and inflation do not: every country uses the same long-run assumptions, which suits a globally diversified portfolio and suits a home-market-heavy one much less. If you hold mostly your own country’s shares, that is a default worth replacing.
• One currency, no exchange-rate risk. Everything is in a single currency; the model doesn’t simulate holding or spending across currencies.

How each risk is simplified

• Investment returns are drawn from a fitted statistical distribution (a normal one, or a fat-tailed one for heavier crashes) calibrated to long-run history. They are not replayed from actual historical sequences, so the model captures the character of market risk (volatility, fat tails, correlation with inflation) but not specific historical episodes.
• Inflation regimes are a stylised two-state model (calm and high), not a forecast of when high inflation will arrive.
• Mortality uses actuarial tables by country, sex, and self-reported health. The shortening of a survivor’s life after losing a partner is applied as a fixed adjustment in the year the partner dies, a simple stand-in for a more detailed bereavement-hazard model. When bereavement and a care episode both hit the same person, the two adjustments are layered one after the other rather than modelled jointly.
• Long-term care is a single onset risk plus a randomly drawn episode length. It doesn’t distinguish levels of care (home help vs. nursing home) beyond the cost figure you provide and the optional lower-cost family-care path. Care does shorten life, since the death hazard is multiplied while an episode is active, but the elevated-hazard window is the episode’s drawn length, the same window the costs are billed over, and recovery restores the normal age curve. The episode length is drawn once per simulated future: if the same future brings a second care episode (or a second unemployment spell, which works the same way), it reuses the first one’s length. That simplification slightly understates how much total care years vary across futures.
• Healthcare costs follow a cost-by-age curve, an approximation of how the bill moves through later life rather than a personalised medical forecast. Fifteen countries carry a curve measured from their own official household surveys, insurers’ records or regulators’ rate tables, and in most of them what retired households actually pay rises gently into the early 80s and then levels off, because fee ceilings and cover priced the same at every age hold the late-life bill down. The United States is the exception in both directions: costs before 65 run several times the level after it, because a retiree there buys their own age-priced cover until Medicare begins (the model prices the fall at 65 and the government help with the premium before it, including the income cap where one extra dollar removes the whole year’s help), and costs after 65 keep climbing, because the top-up cover most American retirees buy alongside Medicare is priced by age at renewal in most policies. The five countries where no official source measures the late-life shape (Norway, Sweden, Italy, Belgium and New Zealand) share a fallback curve, and the reasons each measurement failed are recorded in their data files.
• Rental income is deterministic. Vacancy is a constant occupancy haircut and maintenance is an expected annual cost (the chance of a repair times its size in months of rent). There are no simulated vacancy spells or repair shocks. Over a multi-decade horizon these would largely average out anyway; the risk they’d add is negligible next to markets, inflation, and longevity.
• Some costs use rule-of-thumb growth. Home insurance, for example, is assumed to rise about a percent a year above general inflation. These are reasonable defaults, not researched figures for your specific situation.
• The home-equity fallback is deliberately crude. The home sells in one piece, once, at the brink of insolvency: no downsizing, no reverse mortgage, no buying back in. Its value follows a deterministic drift (no housing-market crashes), the sale takes a flat haircut, the mortgage payoff is estimated from the remaining payments at an assumed rate, and the post-sale rent derives from the home’s value today at a country-typical yield. Proceeds join the portfolio and are then taxed like anything else in it, which is conservative: most countries treat the sale of your own home more kindly than that.
• The investment glidepath follows the first person’s age in a couple, a standard simplification that has a minor effect on the result.

Limits of the premium analyses

The premium diagnostics are built for direction, not decimal precision, and two of them run on a streamlined version of the model to stay fast enough for the browser.

• Sensitivity and counterfactual ranking use a single-person approximation for couples. Trust the ranking of which inputs matter most; treat the exact magnitudes as approximate. (When a counterfactual is actually run, it uses the full model, couples included.)
• The failure-cause breakdown lumps two effects together under “market returns”: ordinary volatility and the unlucky ordering of returns. It doesn’t separate them.
• The sustainable-spending figure is found by search and reported to within about €25/month, so it is close but not exact to the euro.

Reading your results with all this in mind

None of the above means the numbers aren’t useful. It means they’re a well-calibrated guide, not a guarantee. A few habits that follow from it:

1. Watch the range, not the point. A 90% success probability with a small typical shortfall is a very different plan from 90% with a large one. The downside section is where the real information is.
2. Replace defaults with your reality. Every default you override (your actual pension, your actual spending, where your savings sit) makes the result more yours and less generic.
3. Re-run as life changes. A projection is a snapshot of today’s assumptions. It ages; refresh it.