Monthly product demand
Eight years (96 months, 2017–2024) of monthly unit demand for three products sold by a fictional household-goods distributor, Saffron Wharf Supply: ceiling fans (summer peak), water heaters (winter peak) and school backpacks (back-to-school peak in August). Each row is one product-month. The series combine trend, 12-month seasonality and noise, and the water-heater series has a structural break. Built for decomposition, forecasting and honest hold-out evaluation.
- Rows
- 288
- Columns
- 4
- One row is
- one product-month
- Version
- v1
Data dictionary
| Column | Type | Unit | Summary | Description |
|---|---|---|---|---|
monthMonth | date | First day of the month. | ||
month_indexMonth index | integer | mean 48.5, median 48.5, SD 27.76, range 1–96 | Months since the start of the series (1 = January 2017). | |
productProduct | category | 3 levels: Ceiling fan, Water heater, School backpack | Product line. | |
unitsUnits demanded | integer | units | mean 1749, median 1303, SD 1312, range 372–5929 | Units ordered by retailers in the month. |
What it is good for
- Plot and decompose a monthly series into trend, seasonal and irregular parts.
- Fit seasonal naive, moving-average, exponential smoothing and regression-with-season forecasts.
- Evaluate forecasts on a hold-out year with MAE, RMSE and MAPE.
- Detect a structural break and explain why models fitted across it forecast badly.
Methods: time series plot, classical decomposition, seasonal naive forecast, exponential smoothing, Holt–Winters, regression with seasonal dummies, train/test split, MAPE.
Data quality notes
- Water heater demand drops by about 25% from January 2022 (month 61) — a structural break, not noise.
- Seasonal patterns differ in shape and timing between products.
- Long format: filter by product before fitting a single-series model.
- No missing values.
Provenance and licence
Units = level × trend × seasonal index × log-normal noise (sd 0.06–0.08), rounded to whole units. Ceiling fans: level 2 400, +4% per year, peak June–August (index up to ≈1.8). Water heaters: level 1 500, +2% per year, peak December–February; from month 61 (January 2022) the level drops by 25% and stays lower (structural break). Backpacks: level 900, flat, sharp August peak (≈3.5×) with a smaller January bump.
Synthetic data · CC0-1.0 (synthetic data generated by Statistics Quest) · generated deterministically (seed 20241012). It does not describe real people or organisations; any resemblance is coincidental.
Preview
Dataset
Monthly product demand
288 rows · one row = one product-month · synthetic data
| Month | Month index | Product | Units demanded (units) |
|---|---|---|---|
| 2017-01-01 | 1 | Ceiling fan | 1232 |
| 2017-01-01 | 1 | Water heater | 2430 |
| 2017-01-01 | 1 | School backpack | 1080 |
| 2017-02-01 | 2 | Ceiling fan | 1481 |
| 2017-02-01 | 2 | Water heater | 2264 |
| 2017-02-01 | 2 | School backpack | 429 |
| 2017-03-01 | 3 | Ceiling fan | 1948 |
| 2017-03-01 | 3 | Water heater | 1483 |
| 2017-03-01 | 3 | School backpack | 453 |
| 2017-04-01 | 4 | Ceiling fan | 2427 |
| 2017-04-01 | 4 | Water heater | 1304 |
| 2017-04-01 | 4 | School backpack | 539 |
Used in
- Choosing the right chart · Visualising and Exploring Data
- Line charts and time · Visualising and Exploring Data
- Misleading charts · Visualising and Exploring Data
- Time-indexed data · Time Series and Forecasting
- Trend, seasonality and cycles · Time Series and Forecasting
- Moving averages · Time Series and Forecasting
- Simple exponential smoothing · Time Series and Forecasting
- Holt and Holt–Winters · Time Series and Forecasting
- Measuring forecast error · Time Series and Forecasting
- Holdouts and time-series cross-validation · Time Series and Forecasting
- Decomposition · Time Series and Forecasting
- Autocorrelation · Time Series and Forecasting
- Stationarity and autoregressive models · Time Series and Forecasting
- Forecast uncertainty and limits · Time Series and Forecasting
- Demand forecasting for planning · Capstone