Weather adjustment of demand
By weeks builds the link of category sales with the temperature deviation from the norm (a least squares line), estimates its reliability and recalculates the forecast for the expected temperature deviation.
At a glance
- What question does it answer?
- How much weekly sales depend on the temperature deviation from the norm and what adjustment to make to the forecast.
- What data is needed?
- Category sales by week.
- How to read the result?
- Apply the adjustment only at a sufficient R² and with a deviation inside the observed range. Example on the default values: Sales at the temperature norm a, units: 100.95; Slope b, units per °C: −9.57; Reliability of the link R²: 0.988; p-value of the slope: <0. The figures are illustrative: enter your own to see your result.
Why and how to reason
Sales of outerwear and footwear depend on the weather and the link can be measured. In the example there are eight weeks: sales at the norm are 100.9 units, the slope is −9.57 units per degree, R² 0.988; at a deviation of −3 °C the adjustment is +28.7 units, which is 28.4% of the norm.
When to use it
Apply to the weekly forecast of weather-sensitive categories for one region. A general regression is in ols, here a ready scenario with a norm and a threshold.
How to read the result
Apply the adjustment only at a sufficient R² and with a deviation inside the observed range.
Common mistakes
The slope is carried between regions, a small number of weeks is used, weather is mixed with promotions.
What data is needed?
- Category sales by week.
- The temperature deviation from the norm by week.
- The expected deviation of the week.
Inputs
| Field | Unit | Value in the example | Meaning |
|---|---|---|---|
| Expected temperature deviation, °C | °C | −3 | By how many degrees the expected temperature of the week is above the norm (minus: colder). From the weekly weather forecast. The example value is illustrative. |
| Minimum reliability of the link R² | × | 0.3 | The adjustment is applied if R² is not below the threshold. Below it the link is too weak. The example value is illustrative. |
How it is calculated
| Step | Formula |
|---|---|
| Model | sales = a + b × (temperature − norm) |
| Slope | b = sum((x − x̄)(y − ȳ)) ÷ sum((x − x̄)²) over weekly pairs |
| Reliability | R² = 1 − residual sum of squares ÷ total; the p-value from the t-statistic of the slope |
| Adjustment | b × expected deviation = −9.57 × −3 = 28.7; applied at R² not below 0.3 |
Worked numbers
Default values from the app; in the interactive version you enter your own.
- Sales at the temperature norm a, units
- 100.95
- Slope b, units per °C
- −9.57
- Reliability of the link R²
- 0.988
- p-value of the slope
- <0
- Calculated forecast adjustment, units
- +28.7
- Applied adjustment, units
- +28.7
- Adjustment to the norm, %
- +28.4 %
- Observations
- 8
- Link line
- Sales
Data as a table
| Measure | Link line | Sales |
|---|---|---|
| -5 | 149 | 150 |
| -4 | 139 | 140 |
| -2 | 120 | 120 |
| -1 | 111 | 105 |
| 0 | 101 | 100 |
| 1 | 91.4 | 95 |
| 2 | 81.8 | 85 |
| 3 | 72.2 | 70 |
The slope shows a link, not a cause: promotions and holidays in the same weeks mix with the weather effect. For outerwear, footwear and seasonal accessories there is a link; count regional series (Moscow, Siberia) separately.
What to do with the result: add the adjustment to the weekly forecast (reforecast, wosreceipts) only if R² is above the threshold and the deviation is within the observed range; a weekly weather forecast is more reliable than a monthly one.
A typical mistake: carrying the slope between regions, counting on a small number of weeks and mixing the weather effect with promotions.
Nuances and limits
- Carrying the slope between regions.
- Counting on a small number of weeks.
- Mixing weather with promotions.