Free Tool — No Sign-Up
Forecast Accuracy Scorecard: WMAPE, Bias and Tracking Signal
A forecast you never score cannot get better. Accuracy tells you how far off you were; bias tells you which way. Together they show whether the next forecast needs a better guess or a correction.
Paste or upload your forecasts and what actually happened, by line and period. The scorecard gives accuracy (100% minus WMAPE), MAPE, bias and a tracking signal for the whole set, then breaks them out by line and by period so you can see where the misses concentrate. The example rows are illustrative; replace them with yours.
Your forecast and actual rows
Your data never leaves your browser. The rows are analysed on this page. Nothing is uploaded, sent or stored.
Your results
Forecast accuracy (100% minus WMAPE)
95.0%
Weighted MAPE 5.0%: total absolute error divided by total actuals, across 12 rows. Larger rows count for more.
Bias (net error as % of actuals)
+3.0%
The forecast ran above actual. 8 rows over, 4 under. Tracking signal 7.3. The tracking signal is beyond the usual plus or minus 4 rule of thumb, so the misses lean one way. That is bias, not noise.
MAPE (simple average of row errors)
4.1%
Small rows can swing this figure, which is why WMAPE leads.
By line, largest absolute error first
| Line | WMAPE | Bias | Abs error |
|---|---|---|---|
| Product A | 6.9% | +6.9% | $112,000 |
| Product B | 4.1% | -2.0% | $36,000 |
| Service | 1.5% | +0.2% | $9,000 |
By period, in the order given
| Period | WMAPE | Bias |
|---|---|---|
| Jan | 5.0% | +0.6% |
| Feb | 4.5% | +4.5% |
| Mar | 5.9% | +2.4% |
| Apr | 4.6% | +4.6% |
A bias that keeps the same sign period after period is the sign of a forecast that needs a correction, not a better guess.
Illustrative example values are prefilled; replace them with your own. Everything is calculated in your browser, and Performis does not store what you enter. Our analytics record which calculator was used and a few summary figures, such as a rounded result.
The formulas
- Error
- Forecast − Actual (positive = forecast above actual)
- WMAPE
- Σ |Forecast − Actual| ÷ Σ |Actual| × 100
- Forecast accuracy
- 100% − WMAPE (shown as zero if WMAPE is above 100%)
- MAPE
- Average of |Forecast − Actual| ÷ |Actual| × 100, rows with a zero actual left out
- Bias
- Σ (Forecast − Actual) ÷ Σ Actual × 100
- Tracking signal
- Σ (Forecast − Actual) ÷ mean |Forecast − Actual|
How to interpret your result
Read bias before accuracy
A forecast with 8% accuracy loss and no bias is noisy; add detail or shorten the horizon. A forecast with the same loss and a steady lean is mis-set; correct the lean and the accuracy improves without any new insight.
Look at where the error sits
A few lines usually carry most of the absolute error. The by-line table is sorted by absolute dollars so the first rows are where review time pays back. A line with high percentage error and tiny dollars can wait.
Watch the sign by period
If the bias column keeps the same sign across periods, the forecasting routine has a habit, such as optimistic timing of orders or receipts. Name the habit, apply a correction, and score again next cycle to see whether it held.
Do not chase a number in isolation
There is no universal good score; it depends on the line, the lead time and how lumpy the thing is. Use your own history as the yardstick: is this cycle better than the last four?
Write the miss down
The score matters only if something changes. For each large miss, record the cause and what you will do differently, so the next forecast starts from it.
Frequently asked questions
How do you calculate forecast accuracy?
Add up the absolute error on every row (forecast minus actual, ignoring the sign), divide by the total of the actuals, and subtract the result from 100%. That is 100% minus WMAPE. For forecasts of 100, 110 and 90 against actuals of 100 each, the errors are 0, 10 and 10, so WMAPE is 20 ÷ 300 = 6.7% and accuracy is 93.3%.
What is the forecast bias formula?
Forecast bias = (Σ forecast − Σ actual) ÷ Σ actual × 100. A positive result means the forecast ran above actual over the period, a negative one below. For forecasts totalling 3,420 against actuals totalling 3,220, bias is 200 ÷ 3,220 = +6.2%.
How do I calculate forecast accuracy and bias in Excel?
With forecasts in B2:B13 and actuals in C2:C13: bias is =(SUM(B2:B13)-SUM(C2:C13))/SUM(C2:C13), WMAPE is =SUMPRODUCT(ABS(B2:B13-C2:C13))/SUMPRODUCT(ABS(C2:C13)), and accuracy is =1-WMAPE. This tool does the same on pasted rows and also breaks the result out by line and by period.
What is the difference between MAPE and WMAPE?
MAPE averages each row’s percentage error, so a miss on a small line counts as much as a miss on a large one, and a zero actual cannot be used at all. WMAPE divides total error by total actuals, so large rows weigh more and zeros are fine. This tool leads with WMAPE and shows MAPE beside it.
What is forecast bias, and why does it matter more than accuracy?
Bias is the net error as a share of actuals. Over and under misses cancel in bias but not in accuracy. A forecast that is wrong by 5% in random directions is noisy; one that is 5% high every month is leaning, and a lean can be corrected once it is seen. The same accuracy score can hide very different behaviour.
What is a tracking signal?
The cumulative error divided by the mean absolute error. A value that keeps growing in one direction means the misses lean one way. A common rule of thumb treats values beyond about plus or minus 4 as bias worth acting on. It is a prompt to look, not a verdict.
Which forecast should I score?
The one you committed to before the period started, not a reforecast made with the result in sight. Scoring a revised forecast measures how well you can read last month, not how well you can see the next one. Score at the lead time your decisions depend on, such as four weeks ahead for cash.
Is my data stored or sent anywhere?
No. The scorecard runs entirely in your browser, collects no personal information, and does not save what you type.
Go deeper
- Forecast bias vs. forecast accuracy: why a good score can hide a lean
- Rolling forecast vs. annual budget: why forecasts stall
- Variance analysis narrative: writing the why
- Pipeline coverage and forecast accuracy
- Scenario planning with driver-based models
- Operating routines: the weekly, monthly and quarterly cycles
Want the misses explained, not just scored?
Launch partners bring a number like this one and agree it in writing. We build the team that moves it, and measure the result against that number.
Start now