Your 'savings' were never one number

Open your savings account and you get a number. Say it reads 155 000. (Read every figure here in whatever currency you keep — none of this depends on which.)

Now answer a question about it: can I spend 30 000 of this?

You can’t, from the number. Not because 30 000 is more than 155 000, but because the number isn’t the thing you’d need to know. To answer honestly you have to do what you probably already do — open a spreadsheet, or a note on your phone, or just stare at the ceiling for a minute — and write out the list the number is hiding.

The list

Mine, when I first wrote it out properly, looked roughly like this:

• Emergency buffer — 60 000. No date. Its whole job is to be intact on a day I can’t predict.
• Winter tyres and the service — 9 000. About eight weeks away, and not movable.
• Trip next summer — 24 000. Ten months out.
• Kitchen — 55 000. Four years out, give or take a year.

That’s 148 000 promised, against 155 000 held. So the honest answer to can I spend 30 000 is: no. There is 7 000 that isn’t promised to anything, and the other 148 000 has a job. You could still spend the 30 000 — it’s your money — but only by deciding which of those four gets less, and the number on the screen was never going to tell you that. It said 155 000, which is true, and which answers no question you actually had.

This is the whole of zero-based budgeting, and if you’ve used YNAB or anything like it, none of it is news. Give every unit of money a job; what’s left over is what’s actually free. The list above is just that idea applied to a savings account instead of a current account.

But look at the list again, because there’s a second thing in it that envelope budgeting doesn’t touch.

The dates are doing something

Read the four lines and ignore the amounts. What’s left is four dates that have nothing to do with each other: any day now, eight weeks, ten months, four years.

Those dates aren’t decoration. Each one is an instruction about where its money should physically sit.

The buffer has to be worth exactly what it says on the day I need it, which rules out anything that can be down eleven percent that morning. Same for the tyres — eight weeks is not enough time to recover from anything. The trip could tolerate a little movement, though not much, and I’d be irritated rather than stuck. The kitchen money is going to sit somewhere for four years, and putting it in the same place as the tyre money means deliberately accepting nothing for four years in exchange for a safety it doesn’t need.

So one account cannot be right for all four. Not because of anything clever — just because the requirements genuinely conflict. And the moment you accept that, you’re going to do the obvious thing.

What everyone does next, and why it stops working

You split it. The buffer and the near-term things stay in the savings account. The kitchen money goes somewhere it can grow, or at least somewhere it isn’t guaranteed to shrink in real terms. Maybe the trip money goes to a third place. Reasonable, and almost everyone with a bit of money and a bit of patience ends up here.

Then two things happen.

The mapping becomes yours to maintain. Which account holds which promise is now a fact that lives only in your head or in a file. It was implicit before, when there was one account; now it’s bookkeeping, and it goes stale in exactly the way bookkeeping does. Six months on you’re looking at two balances and reconstructing from memory which parts of them belong to what.

The invested part stops matching its promise. You promised the kitchen 55 000. The holding you put it in is worth 53 200 today. Both numbers are true, and neither is wrong. But your list says one thing and your accounts say another, and the difference is a real fact about where you stand — it has just been quietly evicted from your model, because your model was a list of promises and had nowhere to record what you actually hold.

So you round it. Or you keep a “buffer” line that soaks up the difference. Or you top the account up by hand each month so the two agree again. All three are the same move: a person doing arithmetic that a model should be doing.

Two facts about every amount, not one

The thing that makes this tractable is small and slightly annoying, in the way most useful ideas are. It’s this: what money is for and when it’s needed are two separate facts, and every amount has both.

Duration calls the first a bucket — the purpose. The buffer, the tyres, the trip, the kitchen. It calls the second a horizon — when you need it, which is also what decides where it may be held. Near horizons are backed by accounts whose value stays put; far horizons by accounts allowed to move.

Two axes rather than one, and the reason it’s worth the extra column is that the questions you were answering by hand become things you can simply read:

• What’s actually free? The total you hold, minus everything promised. 7 000, in the example above — computed, not remembered, and it moves the instant you promise something new.
• What’s promised inside this account? One invested account can back several buckets at once. The trip, the kitchen and a new bike four years out can all sit inside one holding, each with its own balance and its own target, and you never have to split the holding to see them.
• What has fallen out of step? The gap between what you’ve allocated and what your accounts are funded with — the 55 000 against the 53 200 — is a first-class number. Duration calls it drift, computes it continuously, and shows it rather than absorbing it.
• What should move, and when? As the kitchen gets closer, its money belongs somewhere steadier. That’s a decision with a date attached, and the model knows the date.

And the monthly chore — looking at how much cash is sitting above what the near-term promises actually need, and moving the excess up — is a sweep: proposed when it’s worth doing, with the amount worked out, and executed only if you say so.

What this doesn’t do

It won’t make you money. Duration holds no view on markets, recommends no allocation and moves nothing on your behalf; it records what you decided and shows you the consequences, including the uncomfortable ones.

It isn’t an argument for investing money you’re about to spend, either — that’s precisely what the horizon axis exists to prevent. A repair fund due in March belongs in cash, and the model is what makes that explicit instead of something you re-derive each time you look at a balance.

And it doesn’t model tax. Moving money between horizons can have tax consequences depending on where you live and what the money is held in; Duration neither computes nor warns about them. That’s a real gap and we’d rather state it than bury it — we’re not in a position to do tax with the precision it deserves, and doing it badly would be worse than leaving it out.

The balance was never the answer

None of this is exotic. It’s the observation that a savings account balance is a sum over things that are not alike, and that the moment you write down what those things are and when you need them, you’ve described a structure the account itself cannot hold.

Most people who get this far end up maintaining that structure by hand, in a file, once a month, until they stop. Duration is what happens if you decide the structure is the actual model and the balance is just one of its outputs. It’s in invite-only use while we build it; if the list at the top looked familiar, the concept sets out the whole of it.