Forecast bias
Forecast bias shows whether the demand forecast runs systematically above or below actual sales, as total error divided by total actual sales.
Also calledbias, forecast error direction, tahmin sapması
Formula
Σ (Forecast − Actual) ÷ Σ Actual × 100- Forecast
- The period forecast that was live at order time.
- Actual
- Actual sales for the same period; on stockout days, corrected demand.
- Sign convention
- On this page a positive value means over-forecasting and a negative value under-forecasting. Some sources use the reverse; the report states its convention.
Calculator
Enter your own figures and the result updates instantly. The defaults are the figures from the worked example below.
Result%8
Worked example
How to read it, and what good looks like
Bias and accuracy answer different questions. MAPE measures how big the error is; bias measures its direction. A forecast with large but scattered errors has a bias near zero, and safety stock absorbs the problem. A forecast with small errors that always point the same way cannot be fixed with safety stock: positive bias comes back as days of stock and markdowns, negative bias as lost sales.
Whether bias is systematic is decided with the tracking signal: Σ(Actual − Forecast) ÷ Mean absolute error. The widely used industry control limit is ±4; a series outside it needs intervention. The cause is usually the data rather than the model: stockout days recorded as zero demand, promotion weeks mixed into baseline demand, or a sales target used in place of the forecast. Check those three before changing the model.
Calculate bias weekly and read it by category at the monthly sales and operations planning meeting. In Mindpera, the demand signal feeds Mind Send's replenishment and transfer proposals and Mind Stock's stock requirement; a category whose bias breaches the control limit lands on the planner's task list and is followed until closed. Where the model forecast and the planner's override are not stored in separate fields, installation separates them; only then can you tell whether overrides improve accuracy.
| Context | Value | Note |
|---|---|---|
| Tracking signal control limit | ±4 | Widely used industry rule of thumb; not based on a single study. |
Where it misleads
- At total level bias comes out near zero because one category's over-forecast cancels another's under-forecast. Measure bias separately by category and SKU.
- The sign convention is not written down; the same +8% reads as over-forecasting in one report and under-forecasting in another.
- A sales target is used in place of the forecast; bias stays permanently positive and stock builds a little more every period.
Data you need
- Forecast and actual sales by period and SKU
- Model forecast and planner override in separate fields
- Stockout day and promotion flags
Mindpera products that calculate this KPI
Mind Send
Make your shipments manageable. Depot-to-branch dispatch, branch-to-branch transfer and the initial allocation decision are produced together with the demand forecast.
Mind Stock
Stop treating stock accuracy as something that waits for a count and a month-end close. See movements in and out, expiry dates and profitability on one screen.
Frequently asked questions
- What is forecast bias?
- Forecast bias is a demand forecast running systematically above or below actual sales. It is calculated as the difference between total forecast and total sales, divided by total sales. A positive value means over-forecasting and stock build-up; a negative value means under-forecasting and stockouts. Unlike accuracy, it measures the direction of the error, not its size.
- How is the tracking signal calculated?
- Divide the signed sum of period errors by the mean absolute error: Σ(Actual − Forecast) ÷ Mean absolute error. If the signed error over twelve weeks totals −400 units and the weekly mean absolute error is 60 units, the signal is −6.7. A signal outside the widely used ±4 limit shows the bias is systematic, not chance.
- What is the difference between bias and MAPE?
- MAPE measures the size of the error and ignores direction; bias measures direction and says little about size on its own. A series that misses by 20% a week, over one week and under the next, has a bias near zero. A series that misses by only 5% but always over-forecasts has a positive bias and builds stock. Track both together.