How Accurate Is a Financial Forecast, Really?

The usual objection to a financial forecast is a fair one: nobody can predict the future, so what is the point? Most of what a household forecast does is not prediction. The useful question is which parts are arithmetic, which are assumptions and which are simply out of date, because each fails in a different way and is fixed in a different way.

Three kinds of number

• Arithmetic. Income tax and National Insurance on a given salary, the interest and capital split on a repayment mortgage, a student loan deduction. Given the right inputs these are either correct or a bug, and you can check them against a payslip or a lender's statement.
• Assumptions. Investment returns, house price growth, inflation, the pay rise you expect. These are the genuinely uncertain numbers, and no model, CrestCast included, can remove that.
• Inputs. What you typed in: salary, bills, balances, the date a fixed rate ends. They are facts on the day you enter them and gradually stop being true.

What an assumption is worth over ten years

Take a pot of £20,000 with £500 added every month for ten years. That is £80,000 put in. Grown at 3%, 5% and 7% a year, it finishes in three different places.

At year ten the pot is £96,858, £110,581 or £126,736. A four point gap in the assumed return is worth £29,878, about 37% of everything put in. Nobody can tell you which rate is right, which is why one number from a forecast is best read as one point in a range. Over a decade the range is wide but bounded, and all three paths say the same thing: keep saving. Stretch it much further and the range stops being bounded, which is the subject of why CrestCast forecasts ten years, not thirty.

What an input is worth

Inputs fail differently, because a wrong input is an error rather than uncertainty. A bill of £100 a month missing from the forecast is £12,000 of spending over ten years that the model does not know about. Had that £100 been saved at 5% instead, it would have grown to £15,528. Neither figure is a prediction. Both are the cost of one stale or missing line, and both are fixed in a minute.

That asymmetry is the practical point. You cannot make next year's investment return more certain. You can make sure the salary, the bills and the fixed-rate end date in the model are the ones that apply today, and that is where most of the accuracy a household forecast can have is won or lost.

Where CrestCast focuses

CrestCast is built to get the mechanical category exactly right: income tax on either the rest-of-UK bands or the six Scottish ones, whichever applies to you, National Insurance, all five student loan plans, pension contributions, and your specific loan repayment schedules. Given accurate inputs, the monthly figures match reality closely, because that part genuinely is just arithmetic done correctly, applied to your actual numbers.

Inflation and growth are settings you can see and change, and you can save a plan as a named version and set it beside another with different assumptions, so the range is something you look at rather than imagine.

Inputs stay current because you can log what actually happened when it happens: the forecast carries on from that figure rather than from the one you typed last month.

How to read a forecast

• Treat the first year or two as the sharpest part. Most of it is dated fact: fixed rates, loan payments, known bills.
• Treat later years as a direction, not a promise. Put two versions side by side with different assumptions and look at where they agree.
• Re-check the inputs when something changes: a pay rise, a new fixed rate, a bill that moves, or the tax year changing in April.

The same arithmetic runs on your own pot in the compound interest calculator, and the monthly contribution you would need to reach a target is in the savings goal calculator.

The mechanical side, with tax, National Insurance and mortgage amortisation calculated on real numbers, is worked through in The Three-Way Financial Model, Explained.

Where an AI fits, and where it should not, is covered in what an AI should do with your finances.

How a forecast differs from a budget, with one household worked both ways, is on forecasting vs budgeting.