I retired 20,000 people

Every conversation about drawdown ends up at the same place. What percentage can you safely take out? Four per cent, say the Americans. Three and a half, say the cautious. Whatever the gilt yield is, say the annuity people.

I have written about why I think that question is poorly posed. What I had never done is test it properly, so I did.

What I actually did

I generated 20,000 British households. Pension pots from £60,000 to £1.6m, weighted so most are modest and a few are large, which is roughly how the country looks. Six in ten are couples. A quarter have some defined benefit pension left. Most have a State Pension somewhere between £8,000 and the full amount. Savings outside the pension for about two thirds of them. Spending anchored to what they hold, with a lot of spread either side, because people are not sensible in a uniform way.

Then I gave each household exactly one retirement. Not a thousand simulated futures each: one life, with its own run of market returns and its own inflation, ending at the age they had told the model to plan for. Some got a good decade to start. Some got 2008.

Of the 20,000, about a third ran out of money before the age they were planning for. The median household that failed first fell short at 81. The earliest was 59.

The number everyone argues about

Here is how they did, sorted by the thing everybody discusses: what they spent in year one as a share of everything they had.

It climbs, then collapses, then climbs again. The households spending eight to twelve per cent of everything they owned ran out less than half as often as the ones spending six to eight. That is not noise: it held in every population I generated.

Stare at that for a moment, because it is the whole article. If the withdrawal rate were the thing that mattered, that line would climb steadily from top to bottom. Instead it wanders. A measure that wanders is not measuring the thing that decides the outcome.

Why the line falls over

Because a percentage of your savings ignores the money that turns up whether your investments behave or not.

Take a couple with two full State Pensions, around £25,000 between them, spending £30,000 a year, with a £90,000 pot. On the usual measure they are drawing 33% of everything they own, which sounds like a catastrophe. In reality their pot has to find £5,000 a year and the State does the rest. They will very probably be fine.

Now take someone with £600,000, spending £36,000. Six per cent, which sounds punchy but survivable. If they retire at 57 with no State Pension for a decade, that pot is carrying the entire £36,000 on its own for ten years before any help arrives.

Same arithmetic, opposite answers. So I measured the other thing instead.

The number that did predict it

For each household I worked out the part of their spending the pot itself has to fund, once the State Pension and any defined benefit income have done their bit, and measured that against everything they held.

That is a staircase. Every step up, and the odds get worse, with no reversal anywhere. Households whose retirement lasted needed their pot to cover 1.4% a year on average. Households that ran out needed 3.9%.

The same story arrives from the other direction if you ask how much of their spending was already covered by income that arrives regardless.

Cover none of your spending with guaranteed income and, in this model, you have roughly a three in four chance of running out. Cover all of it and you have about one in fourteen.

So what is the better question?

Not "what percentage can I take?" but "how big is the gap between what I want to spend and what turns up anyway?"

It is a more awkward question, because it needs two numbers you might not have to hand: what you will actually spend, and what your State Pension and any old scheme will actually pay. But it is the one the model cares about, and it explains things the percentage cannot. It is why deferring the State Pension, or filling National Insurance gaps, or buying a small annuity, move the odds more than they look like they should. They all shrink the gap.

It also explains why the years before your State Pension arrives are the dangerous ones. For most people the gap is at its widest the day they stop work and narrows from 66 or 67 onwards. The plans that failed were mostly not brought down by extravagance. They were brought down by a wide gap held open for a long time.

What this does not show

I would rather say this myself than have it said to me.

These are invented households, not real ones, and their spending is my guess at how people behave rather than a survey. Everyone here spends the same amount in real terms every year until they die, which nobody does. Nobody adjusts when markets fall, nobody has a cash buffer, nobody buys an annuity, nobody downsizes, and nobody needs a care home. Returns average five per cent with twelve per cent of variation and inflation averages two and a half, which are assumptions rather than forecasts. And "ran out" means the model could not meet the spending in some year before their target age, which in real life is the point where someone spends less, not the point where life stops.

What I think survives all of that is the comparison rather than the level. The third who ran out is an artefact of my assumptions. The shape of the two charts is not: one measure predicts and the other does not, and that held in every population I generated.

If you have been agonising over whether four per cent is too much, you may be optimising the wrong number. Mine is a gap I will be holding open from 58 to 67, which on this evidence is the part of my plan I should be worrying about, and I was not.

The code that generated these households is in the open, like the rest of it. If you think my invented population is unfair, or my assumptions are doing the work, tell me and I will run it your way. Corrections go in with credit, as always.

Method: 20,000 households across five independently generated populations, one simulated life each, run on this site's engine at version 1.58. Assumptions as stated. Nothing here is a recommendation, and I am not a financial adviser.