How Far Ahead Can You Actually Predict Your Bank Balance?

Kia ora. Every forecasting tool will happily draw you a line thirty years out. The line is smooth, it slopes upward, and it is almost entirely fiction.

Worth knowing where the fiction starts.

Two very different things get called "forecasting"

Projecting what you already know. Your rent is $430 a week. Your pay lands on Thursday. Your car insurance renews on 14 March for roughly what it was last year. None of that is a prediction so much as arithmetic on facts you already have.

Modelling what you don't. What your income will be in 2031, whether interest rates hold, whether you'll have another child, whether the industry you work in still exists.

The first is a calculation. The second is a scenario with an air of authority. Both get drawn as the same confident blue line, and that's the problem.

Where the reliability actually falls off

Accuracy doesn't stop dead at a date. It decays, and it decays at different speeds for different things.

Next 7 days — very high. Everything material has either happened or been scheduled. Direct debits are queued. The only real unknown is discretionary spending, and one week of that has a small range.

8 to 30 days — high. One full bill cycle. Rent, power, phone, subscriptions, one or two pay days. Discretionary spending starts to matter but averages out.

1 to 3 months — good. Most household patterns are stable across a quarter. Quarterly things appear: rates instalments, some insurance, tax for the self-employed.

3 to 12 months — useful, with a widening range. You capture the seasonal shape. Christmas. The February back-to-school hit. Winter power bills that run 40 to 60 percent above summer in an uninsulated NZ house. Annual renewals. This is where a forecast stops answering "will I make it" and starts answering "what have I forgotten".

Past 12 months — you've left the data. Beyond a year you're compounding assumptions about income, inflation, rates and life. Each is uncertain, and they multiply rather than add.

Why a year is the natural stopping point

A year is one complete seasonal cycle. Every annual bill you have lands exactly once inside it. Rates instalments, all four. One Christmas. One winter. One birthday season.

Run 11 months and you might miss the rego. Run 18 and you've counted one Christmas twice with nothing to check the second against.

That's the real argument for a 12-month horizon, and it isn't a technical limit. It's that a year is the shortest window that contains everything, and the longest one built from your actual history rather than assumptions about your future.

What "confidence" should mean on a chart

A forecast that shows one line is hiding something. Any honest projection is a range that widens with distance — narrow next week, wide next spring.

If a tool shows a single crisp number for July next year, it isn't more accurate than one showing a range. It's less honest.

Steady states its confidence for whatever window you're looking at and lets it fall with distance, because a 340-day-out figure presented with the same certainty as tomorrow's is a lie told with a nice gradient. You can see 7 days, 30 days, 6 months or 1 year, and the further out you go the more the chart admits it's estimating.

The multi-decade projections

Some tools — PocketSmith is the well-known one in New Zealand — will project ten, thirty, sixty years. There is a real use for that: retirement modelling, mortgage scenarios, "what if I salary-sacrifice another 3 percent".

But be clear about what you're reading. A 30-year projection isn't a forecast of your balance. It's a compound-interest calculation on a set of assumptions you supplied, and its accuracy is entirely the accuracy of those assumptions. Change the assumed return from 5 percent to 4 percent and the 2056 figure moves by a six-figure sum.

Useful for comparing two decisions. Not useful for deciding whether to buy the shoes.

What to actually ask a forecast

Not "what will my balance be". Ask:

• Do I run out before payday? Days, not months.
• Which upcoming week is the tight one? Where bills cluster.
• What lands this quarter I've forgotten? Rates, rego, renewals.
• Can I afford this specific thing? A one-off against a known window.
• Am I trending up or down? Direction, over months, is more reliable than any single future number.

Every one of those is answerable inside a year from data you already have. None of them needs a 2050 projection.

The short version

Seven days: near certain. A month: reliable. A quarter: good. A year: useful, widening, and the point at which you've seen every recurring cost you have.

Past that you're not forecasting. You're modelling — which is a fine thing to do, as long as nobody's pretending the line is a fact.