FIRE glossary
Historical Backtest
Replaying your plan against one specific real historical sequence of market returns — such as retiring at the start of 1929 — rather than a randomly generated one.
One real sequence, not an average
Ember's historicalBacktest function takes a start year you choose and a horizon in years, then pulls the actual annual returns that followed from Ember's historical dataset — real (inflation-adjusted) equity, bond, cash and FX figures running from 1928 to 2025 — and feeds that exact sequence into the same path simulator Monte Carlo uses. Nothing is sampled or randomised: if you pick 1929, you get the real 1929 crash, the real Depression-era recovery, and everything that actually happened afterwards, year by year, applied to your own contributions, spending and withdrawals.
Why the order of returns is the whole point
A backtest exists to make sequence-of-returns risk concrete rather than abstract. Two starting years with an identical long-run average return can produce very different outcomes for the same portfolio, because a bad run of years early in retirement does more damage than the same bad years arriving later, after the pot has had time to recover. Running your own numbers against a genuinely bad historical start year — not just a good one — is the useful test; picking only flattering years to try defeats the purpose.
What it doesn't claim
A single historical sequence is one real data point, not a distribution — it tells you how your plan would have fared in that specific past, not the odds of a similar sequence recurring. The underlying data blends sources by region and era (UK data from the JST Macrohistory Database, US data from Damodaran/FRED, global equity proxied by US returns before 1970 where no global index existed), and if a requested horizon runs past 2025 the simulation truncates and is labelled as such rather than padded with invented years. Pair a backtest with a Monte Carlo simulation, which samples many possible sequences, to see the fuller range rather than relying on any one historical path — Ember is an information and modelling tool, not a forecast of what will happen next.
Across borders
Ember's historical series has genuinely separate UK and US return and inflation histories, plus GBP/USD (and more recently GBP/EUR) FX data, rather than one blended "global" number — so a backtest run against a GBP-denominated plan and the same plan converted to USD can diverge, not just from currency conversion at the end but from the underlying return and inflation series actually being different across that period. Before 1970, global equity returns are proxied with US data because no global index existed yet, which matters most for plans anchored well outside the US or UK.
Common questions
Is a historical backtest more accurate than Monte Carlo?
Neither is more "accurate" — they answer different questions. A backtest shows exactly what would have happened to your plan in one real past sequence. Monte Carlo samples many possible sequences to show a spread of outcomes. Used together they're more informative than either alone.
Can I backtest any start year?
Only years covered by Ember's historical dataset, which currently runs 1928–2025. Requesting a horizon that runs past the end of the data gets truncated at 2025, and the result says so rather than extending the sequence artificially.
Does a bad historical start year mean my plan will fail the same way?
No — it means your specific plan, applied to that specific real sequence, would have produced that specific result. Future sequences won't repeat the past exactly. Treat a bad backtest year as a stress test of your assumptions, not a prediction.
Related terms
See historical backtest in your numbers
The free calculator gives a rough estimate; the full planner models your actual accounts, pensions, residency moves and taxes — with the maths behind every figure shown.