Only used for labelling the chart. The order of the rows is what matters.
Sales, sign-ups, traffic, cost — whatever you are trying to predict.
Forecast Starter
Time Series — Start Here
Your first time series tool. Look at what a series is actually made of, extend it with a rule you can say out loud, then find out whether that rule beats doing nothing at all. No parameters to guess, no Python to load, nothing to install.
HOW THIS WORKS
A forecast is not a magic box. It is a rule for extending a line, and you can say every rule on this page in one plain sentence. This tool has three steps, in this order.
-
1
See the shape. Split the series into the parts it is made of — its level, its direction, its repeating cycle, and the leftover noise. This is what tells you which rule has a chance.
-
2
Extend the line. Pick a rule and watch it run off the end of the data. Change the rule, change the settings, see the line move.
-
3
Score it honestly. Hide the last few periods, forecast them, and compare against what really happened — scored against the simplest possible rule, "just repeat the previous period".
The one idea to take away: a forecasting method has to beat "same as last period" before it is worth anything. A surprising number of expensive, complicated forecasts do not, and the only way to find out is to test them on data they have never seen.
What this tool deliberately does not do
- It does not use external predictors — ad spend, price, promotions. Every method here works only from the series' own past. When you need that, move up to the ARX tool.
- It does not test for stationarity, difference the series, or fit autoregressive terms. Those are the next rung, not this one.
- It does not produce p-values, because none of these methods has one.
📚
Use a Case Study
Four series, each one teaching a different answer to "which method should I use". Start with the first — it is the one where the cleverest method loses.
📤
Upload Your Data
A CSV with one column of dates or period labels and one column of numbers, in time order, with no gaps.
Drag & drop a CSV (.csv, .tsv, .txt)
First row = column headers. One row per period, oldest first.
No file uploaded.
CHOOSE YOUR COLUMNS
Step 1 SEE THE SHAPE
How many periods before the pattern repeats: 12 for monthly data with a yearly cycle, 7 for daily data with a weekly one, 4 for quarterly.
How the parts combine:
Added means the December bump is the same number of units every year. Multiplied means it is the same percentage, so the bump grows as the business does.
The series itself
Trend — the underlying direction
The series with its wobble and its cycle smoothed away.
Season — the part that repeats
The same shape every cycle, above and below the trend.
Noise — everything left over
Should look structureless. Big spikes here are usually events, not noise.
Step 2 EXTEND THE LINE
Pick a rule. Each one is a single sentence — if you cannot say it out loud, do not use it.
The rule, written out
In symbols
With your numbers
Forecast
Step 3 SCORE IT HONESTLY
Every method looks accurate on the data it was fitted to. The only honest test is to hide the last few periods, forecast them, and compare against what actually happened.
These are cut from the end, never chosen at random — a method that saw next month while predicting this one is not being tested, it is being told.
Watch out: every change you make (a method, a slider, the cycle length) is re-scored on these same hidden periods. Once you pick settings by watching this table, the hidden periods are helping to choose the model, so a score you tuned your way to is optimistic. Check it at a different number of hidden periods before you believe it.
Every method, on the same hidden periods
| Method | Typical miss (MAE) | Typical miss (%) | vs. "repeat last period" |
|---|---|---|---|
| Load data to score the methods. | |||
How to read the last column
- Below 1.00 — better than simply repeating the previous period. The lower the better.
- 1.00 — exactly as good as doing nothing clever.
- Above 1.00 — worse than doing nothing clever. Prefer the naive forecast. If slider-tuning drags the number below 1.00, that is not a rescue: the settings were picked on the same hidden periods that score them.
MAE is the average size of the miss, ignoring whether it was too high or too low. The percentage column is left blank when any hidden period sat at or near zero, because dividing by it would produce a meaningless number.
WHAT THIS MEANS
APA-Style Report
Load data to see the write-up.
Managerial Interpretation
Business-focused interpretation will appear here.