Run a retirement plan through a Monte Carlo simulator and one number ends up larger than everything else on the screen. The ring fills, it reads 85%, and on this site it goes green, because the gauge turns green at 80.
Green, pass, done.
Except 85% is not a pass mark, and it is not an 85% chance of your retirement working. It is the share of a thousand invented futures, built from assumptions you chose yourself, in which the money lasted. Everything worth knowing sits in the detail underneath it.
What the gauge counts
Take a saver of 55 with £1,600,000 invested, planning to spend £45,000 a year in today's money, assuming a 5% real return and this simulator's default 15% volatility. The pound signs are incidental, by the way. The model never converts anything, so identical figures give identical percentages whatever symbol sits in front of them, though the return and volatility worth typing do differ from one market to the next. And the draw is worth naming here: £45,000 on £1,600,000 is 2.81% a year, 1.2 points under the 4% convention.
The Monte Carlo simulator builds 1,000 month-by-month return sequences, runs the plan down each one, and scores a path a success only if it reached its target and the money is still there at 95. This plan scores 84.7%. The gauge rounds that to 85%.
Put it the other way round. In 153 of those 1,000 futures the money ran out at some point over the next forty years, on a draw nearly a third smaller than the one most people treat as safe. That is the entire content of the number. When a path failed, by how much, and what the other 847 looked like: none of it is in there.
Forty years is the second thing hiding in the setup. The horizon in this app is fixed at 95, so a saver of 55 is asking a much harder question than a saver of 65 with the same pot and the same budget. Move the start age and change nothing else, and the identical plan reads 85% from 55 and 92% from 65. Thirty years of drawdown is a different problem from forty, and the gauge quietly prices that in without ever mentioning it.
Most of the failures arrive very late
The simulator records the year each path hit zero, and those years bunch up at the far end. The earliest failure in this plan lands at 72, and it is one path in a thousand. By 80, 36 have run out, under a quarter of all the failures. By 85 it is 74, about half. By 90, 113, three quarters of them. The rest go in the last five years.
Now weigh those ages by the chance of being there to see them. The mortality model behind the Rich, Broke or Dead view approximates contemporary UK and US unisex life tables, and gives a 55-year-old a 57% chance of reaching 80, 39% of reaching 85 and 21% of reaching 90. Push the failure years back through that curve and the chance of actually outliving your money, rather than of a simulated path hitting zero in a year you were never going to see, comes out at 5.75%. The gauge says 15% failed. The question you were asking is closer to the 5.75%.
Treat that as a rough sighting rather than a measurement. The model is unisex, so a woman's odds are better than it says. It is three constants matched to a few anchor ages rather than a real life table, and it errs towards early death, which flatters this particular calculation. Your own country's numbers will differ, sometimes by years, and the Human Mortality Database publishes the real tables to look them up in. For a couple it is the wrong curve altogether, because the money has to last to the second death rather than the first.
None of which makes a late failure harmless. Running out at 88 is a genuine catastrophe, arriving at exactly the age when you can do least about it. But "a 15% chance of ruin" reads the gauge badly wrong.
Inside the 85% is an enormous range
The fan chart around the gauge is drawn across all 1,000 paths, ruined ones included, which is precisely why it is worth looking at. At 85 the tenth percentile path holds £143,644, the median £2,155,599 and the ninetieth £8,689,375. Carry on to 95 and the median is £2,487,868 with the top tenth above £13,855,063, while the tenth percentile is £0, because by then 15% of the paths are broke.
Read that median again. The typical outcome for someone drawing £45,000 a year for forty years is dying with more than they retired on, and a good tenth of these futures finish with nearly nine times the starting pot. A path that ends on £13m and a path that ends on £500 score the same single point. Any number that flattens a spread that wide needs the chart beside it.
The number carries its own noise
Change nothing at all and reseed the random generator, which means going into the engine rather than the page, since the seed is not a control the app offers. Across eight arbitrary seeds the same plan scored anywhere between 83.5% and 87.8%, which is 84 to 88 on the gauge. The run count you can change, and 500 runs gives 87%, 1,000 gives 85%, 5,000 gives 86%.
So read the level loosely. Anything within two or three points of 85% is the same plan as far as this model is concerned.
Comparisons are a different matter, and much sharper than that spread suggests. The seed is pinned, so two versions of a plan are scored against the identical thousand market histories and only the change you made can move the result. Trim the budget by 3.3%, from £45,000 to £43,500, and the score improves on every one of forty seeds tried, by 1.44 points on average. That is well inside the noise of the level and still completely real. Doubt the number on its own; trust the direction when you change one thing.
It moves on assumptions, not on facts
Cutting the budget by a tenth, to £40,500, lifts the score from 84.7% to 88.5%. Dropping the assumed return from 5% to 4% pulls it down to 75%. And then there is the volatility box, editable in the FIRE calculator's Monte Carlo mode, which at 10% reads 97%, at 15% reads 85%, at 18% reads 75%, and at 20% reads 68%.
That ladder is doing something the label never admits to. The 5% you typed is an average of yearly returns, and the simulator scatters each month around it, so the compounded result lands below the average by roughly half the variance. Type 5% and the median path actually compounds at 4.49% a year at 10% volatility, 3.84% at the default 15%, and 2.95% at 20%. Somewhere between a third and a half of the fall from 97% to 68% is that quiet markdown of the growth rate rather than any widening of the spread, depending on which of the two you hold still while you measure the other. Which also means the plan above, the one set up at 5%, is being scored on a median return under 4%.
Worth knowing too that 15% is a mild default. The historical series this app backtests against has a standard deviation of 19.4%, and long-run real equity volatility sits nearer 18% to 20%. Moving that box to 20 is closer to correcting it than to stress testing it.
History gives a very different answer, for a reason
Run the same saver through the historical backtest, which replays recorded returns instead of drawing random ones, and the plan fails in none of its 124 start windows. A hundred per cent.
Which is worth having, though not for the reason it first appears. The backtest never reads your expected-return box. It runs on its own series of approximate US large-cap real returns from 1900, derived from the long price, dividend and earnings record Robert Shiller publishes, and that series averages 8.2% a year, or 6.34% compounded. Feed 6.34% back into the simulator and it scores 92.8%. So about half of the fifteen-point gap is just two tools answering with different growth assumptions, not one of them understanding markets better than the other.
Take the 100% with salt as well. The windows overlap by 39 years out of 40, which makes them nearer three independent lifetimes than 124 trials, and the later ones wrap the series back round to 1900 to fill out the horizon. It is US equity data, rounded, applied here to a plan denominated in pounds. Two tools, two answers, and the honest reading is the range between them.
A perfect score can be bought, and the price is hidden
There is a much faster route to a gauge reading 100%. Leave every figure alone and change only the withdrawal strategy, from fixed spending to one that flexes with the market.
Fixed spending scores 84.7%, with a median retirement income of exactly the £45,000 asked for. A simplified Guyton-Klinger guardrail, which cuts the draw by a tenth after bad markets and raises it by a tenth after good ones, scores 99.9% on a median income of £60,496. The percentage-of-portfolio rule, variable percentage withdrawal, scores a clean 100% and a median of £61,771, and it cannot fail by construction: it always draws a share of whatever is in the pot, so the balance can shrink for decades without ever reaching zero.
Both flexible rules pay the median retiree about a third more, and that is worth understanding before you read it as free money. Neither rule spends your £45,000. They spend the withdrawal-rate box, which still says 4%, and 4% of £1.6m is £64,000. The guardrail sees a draw of 2.81% sitting well under its 4% and raises spending in the first year, then keeps raising it until the two meet. Set the box to the rate you actually intend to draw and most of that extra third goes away. The cost turns up at the other end of the distribution. The tenth-percentile worst year is £13,827 under guardrails and £13,071 under VPW, against £30,395 for fixed spending, which looks like a straightforward argument for fixed spending until you notice what that statistic leaves out. It counts only years in which money was actually drawn. For the 153 paths that ran dry, every year afterwards is zero, and not one of them is in the figure. The fixed plan's real worst year is nothing at all, for the rest of a life.
So that is the trade, in one line. Flexible rules convert a small chance of nothing into an unlucky tenth who take deep cuts instead. A 100% success rate is not a safer plan. It is a different definition of failing, and the gauge cannot tell you which one you would rather live with.
Use it as a dial, not a verdict
The number compares two versions of your own plan honestly, as long as every assumption is held still and you ignore gaps under three points. It stress tests well: push volatility to 20 and the return down a point, then ask whether the answer is still one you would act on. And it frames the two questions it cannot answer itself, which are when the failures land and how small the income gets. Where to draw the acceptable line has a literature of its own, and Cooley, Hubbard and Walz, the authors behind the Trinity study, went at it directly in Portfolio Success Rates: Where to Draw the Line, which is a better guide than any single threshold.
The standalone simulator runs at the fixed defaults, 1,000 paths at 15%, so the box you want is on the FIRE calculator: switch the chart to Monte Carlo and the volatility and run count become editable. Put your own figures in, then change one thing, volatility, from 15 to 20. Whatever that does to the gauge is the most honest thing it will tell you all day. The methodology page sets out exactly what this model assumes, and the 4% rule covers the fixed withdrawal convention the whole thing rests on.