Analytics
Time Series Forecast
Projects a dated measure forward — the next weeks, months or quarters — with Holt-Winters (level, trend, season) or a seasonal naive forecast.
When to use it
Use it for a quick, explainable projection of sales, demand or volumes from their own history.
In a pipeline
File InputTime Series ForecastChart
Settings
| Setting | Type | Required | Default or hint |
|---|---|---|---|
| Date column | select | Yes | Select column… |
| Value column | select | Yes | Select column… |
| Periods ahead | number | — | 12 |
| Method | select | — | holt-winters |
| Season length (0 = from the dates) | number | — | 0 |
| ARIMA order p,d,q | text | — | 1,1,1 |
| Seasonal order P,D,Q | text | — | 0,1,1 |
Example
Next two quarters
- Date column=Quarter
- Value column=Sales
- Periods ahead=2
- Method=Seasonal naive (repeat the last season)
| Quarter | Sales |
|---|---|
| 2025-01-01 | 100 |
| 2026-07-01 | 190 |
| 2026-10-01 | 130 |
| Quarter | Sales | Forecast | Type |
|---|---|---|---|
| 2026-07-01 | 190 | — | Actual |
| 2026-10-01 | 130 | — | Actual |
| 2027-01-01 | — | 110 | Forecast |
| 2027-04-01 | — | 150 | Forecast |
With all eight quarters of 2025–2026, the season is 4 (worked out from the dates), so the forecast repeats the last four quarters.
Worth knowing
- The step is read from the dates: months, quarters or years when the dates fall on the same day of the month, otherwise the typical gap in days (day, week…).
- Season length 0 means automatic: 12 for months, 4 for quarters, 7 for days, 52 for weeks. Too little history for two seasons drops the season and says so.
- Holt-Winters picks its smoothing (α, β, γ) by testing a grid against the history; the note shows the choice.
- ARIMA takes an order p,d,q (AR, MA, ARMA and random walk with drift are special cases: 1,0,0 · 0,0,1 · 1,0,1 · 0,1,0); Seasonal ARIMA adds P,D,Q at the season length; Auto-ARIMA picks d and D by differencing and p, q, P, Q (up to 3, 3, 1, 1) by the lowest AIC. Fitted by conditional least squares in FlowDoo itself.
- Repeated dates are added together; rows without a readable date or number are skipped and counted.