A spending forecast is as accurate as your own history is steady, and for most categories that means it can tell you a likely range, not a single figure. Next month's groceries are probably somewhere between two numbers you can name, and a forecast that says so honestly is more useful than one that quotes a confident total to the dollar. (We build Tobsy, a manual spending tracker that forecasts as a range, so we think about this a lot.)
Why spending is lumpy
Most months are not like the last one. A few reasons show up in nearly everyone's numbers:
- Irregular bills. Car rego, a water bill every quarter, a dentist visit. They land in some months and not others.
- Annual costs. Insurance, a software renewal, school fees. One month carries a whole year.
- Seasonal months. December is the classic: gifts, hosting, travel. For some people it is winter heating, or the month the kids go back to school.
- Ordinary wobble. Five weeks in a month instead of four, a big shop before a long weekend, friends visiting.
So a single number for next month is a guess wearing a suit. It looks precise, and the precision is not real.
The worked example: twelve months of groceries
Here is a realistic year of grocery spending for a household. Eleven ordinary months, and one December when they hosted the family.
| Month | Groceries |
|---|---|
| January | A$540 |
| February | A$610 |
| March | A$685 |
| April | A$575 |
| May | A$650 |
| June | A$720 |
| July | A$560 |
| August | A$630 |
| September | A$595 |
| October | A$700 |
| November | A$620 |
| December | A$1,180 |
The mean is dragged by one month
The mean (the everyday average) adds all twelve and divides by twelve: A$672. Without December it would be A$626. One month moved the "typical" figure by A$46, and A$672 sits above eight of the eleven ordinary months. Used as a forecast for an ordinary January, it would be too high most of the time.
The median barely moves
The median is the middle value once the months are sorted. With all twelve it is A$625 (halfway between A$620 and A$630). Set December aside and it is A$620. The big month moved it by A$5, because the median only cares which values are in the middle, not how far out the edges go.
| With December | Without December | |
|---|---|---|
| Mean | A$672 | A$626 |
| Median | A$625 | A$620 |
A usual range, step by step
A middle value is half the answer. The other half is how much a normal month wobbles around it. A robust way to measure that is to take each month's distance from the median, then take the median of those distances. Statisticians call this the median absolute deviation, or MAD, if you are curious.
- December sits A$555 from the middle, while the other distances are all under A$100. It is far enough out to be set aside as an unusual month, not as part of the pattern.
- The eleven ordinary months have a median of A$620.
- Their distances from A$620 are 0, 10, 10, 25, 30, 45, 60, 65, 80, 80 and 100. The middle one is A$45: that is the typical wobble.
- Widen it to one and a half times: A$67.50. That is roughly one standard deviation's worth, measured so a single odd month cannot stretch it.
- The usual range is A$620 minus and plus A$67.50: about A$553 to A$688.
Eight of the eleven ordinary months land inside it, and the other three sit just outside. That is what a usual range is for: most months, not every month.
Compare a band built the textbook way, the mean plus and minus one standard deviation, on all twelve months: about A$510 to A$834. December has lifted the middle and more than doubled the width, so the band now says very little.
What next month looks like
For an ordinary month ahead, the honest forecast is: likely about A$620, somewhere between A$553 and A$688. If you know a big month is coming, a birthday dinner at home or a December like last year's, add it as a known item on top rather than hoping the average absorbs it.
Why the total is tighter than its parts
If five categories each wobble by about A$50, the month as a whole does not wobble by A$250. For it to land that high, all five would need their expensive month at once, like five coins all landing heads. Usually one runs high while another runs low, so the whole month's range is narrower than the category ranges added up. A good forecast shows both: each category honest about itself, and the month honest about the month.
How much history you need
With one or two months, there is no pattern yet, only sightings. Three months is the least that can draw a range, and even then it is fragile: two similar months make an ordinary third look dramatic. A seasonal month needs more again, at least two years of the same month agreeing before you can say "December runs high for us" with any confidence.
Tobsy follows that pattern: it draws a usual range from three months of a category, judges the fourth month against it, and labels the range as still learning until about five months. Its forecast lifts a month like December only once two years of that month agree.
When another method fits better
If your spending is mostly fixed costs, rent, a loan and a handful of bills at set amounts, you do not need ranges at all: a dated list of what is due, in a spreadsheet or your bank's own scheduled payments, forecasts that precisely. If what you want is a day-by-day projection of an account balance, a tool connected to your accounts that models the balance itself will suit you better than a category forecast. And with under three months of records, the honest forecast is last month's figure plus a generous margin, while the history builds.
Questions
How accurate are budget forecasts?
As accurate as your history is steady. Fixed costs can be forecast to the dollar. Groceries, dining and fuel move from month to month, so a forecast for them is honest only as a range, usually the middle of your recent months plus and minus their typical wobble.
Why use the median instead of the average?
Because one big month drags the average. In the example above, a single December moved the average by A$46 and the median by A$5. The median describes a typical month; the average describes the year's total divided evenly.
What is a usual range?
The band most of your months land in for one category: the middle of your months, plus and minus about one and a half times their typical distance from that middle. It is drawn from your own history, with no outside benchmark.
How many months of spending do I need to predict next month?
Three months can draw a rough range. Five or six make it steadier. A seasonal month such as December needs two years of the same month before it can be predicted with any confidence.
Should I leave the unusual month out?
Out of the usual range, yes, because it describes ordinary months. Keep it in your records, and plan for it separately when you know it is coming.
Sources
- Wikipedia: Median absolute deviation — "the median of the absolute deviations from the data's median"; "more resilient to outliers in a data set than the standard deviation"; the 1.4826 scale factor to a standard deviation for normal data; read 2026-10-19
- Wikipedia: Median — the median "is not skewed by a small proportion of extreme values"; read 2026-10-19
- The worked example's figures are illustrative, calculated for this page.
Pass it on