Forecast Accuracy Calculator
Measure forecast error by item and across a portfolio. MAPE, WMAPE, bias, MAD and tracking signal, with clear treatment of zero actuals.
Actual versus forecast
Paste comma- or semicolon-separated CSV, including the header. Use decimal points without thousands separators, YYYY-MM-DD dates, and one consistent unit per item. Use the shown column names; unrecognized columns are refused. IDs are case-sensitive text: leading zeros are kept. Each paste supports up to 5,000 data rows. Calculations run in your browser; nothing is uploaded.
item, period, actual, forecast. One row per item/period, using a consistent unit and evaluation window. Period labels are kept as text.
Accuracy results
1 items, 2 observations. Positive bias = over-forecasting.
| Item / scope | Periods | MAPE | WMAPE | Bias | MAD | Tracking signal | Notes |
|---|---|---|---|---|---|---|---|
| All items (pooled) | 2 | 10% | 10% | 3.33% | 15 | 0.6667 | — |
| A | 2 | 10% | 10% | 3.33% | 15 | 0.6667 | — |
MAPE is undefined if any actual is zero. WMAPE and bias need positive total actuals. Total metrics pool observation errors; compare items in compatible units.
Formulas and worked example
Let e = forecast − actual and n = number of observations. MAPE = 100 × mean(|e| ÷ actual). WMAPE = 100 × Σ|e| ÷ Σactual. Bias (%) = 100 × Σe ÷ Σactual. MAD = Σ|e| ÷ n. Tracking signal = Σe ÷ MAD (zero for a perfect forecast).
Item A has actuals of 100 and 200, with forecasts of 90 and 220. Errors are −10 and +20; absolute errors sum to 30, signed errors sum to +10, and actuals sum to 300. MAPE = 10%, WMAPE = 10%, bias = +3.3333%, MAD = 15, and tracking signal = +0.6667. The worked example is locked by a calculation test.
Use WMAPE when you want high-volume items to carry more weight, especially when zero or tiny actuals make MAPE unusable. Use MAD to read error in the original units. Positive bias signals excess forecasts that may create excess stock; negative bias can leave too little supply. A low signed bias can still hide large errors in both directions, so read it alongside WMAPE or MAD.
FAQ
When should I use WMAPE instead of MAPE?
MAPE gives every item-period percentage equal weight, so tiny actual volumes can dominate. Use WMAPE for a volume-weighted portfolio view when items share comparable units. WMAPE keeps errors from zero-actual periods and divides by total actual volume. It is undefined when that total is zero.
Why is MAPE undefined when actual is zero?
The percentage error for that observation divides by zero. This tool marks MAPE undefined for any item or total containing a zero actual, rather than silently dropping those observations. MAD remains available, and WMAPE is available if total actual volume is positive.
How do I read bias and tracking signal?
Error here is forecast minus actual. Positive bias means over-forecasting; negative means under-forecasting. Tracking signal is cumulative signed error divided by MAD, with the same sign. A perfect forecast has tracking signal zero. Thresholds such as ±4 are screening conventions, not universal acceptance limits; evaluate them over a consistent monitoring window.
How are totals calculated?
All item-period errors are pooled before calculating total metrics. Absolute errors are added before dividing, so opposite errors do not cancel in WMAPE or MAD. Use comparable units and the same evaluation window across items. Duplicate item/period pairs and negative values are refused.
Is my data uploaded?
No. Calculations and CSV parsing run in your browser, and the numbers and files you enter are never sent to a server, unless you choose to share a link, which carries the single-item inputs in its address. The page has no ads, sign-up or third-party trackers; the site host keeps standard anonymous visit counts.
Related tools
Want the full picture for all your SKUs?
A fixed-price inventory health and safety-stock checkup: send one export and get a workbook with ABC/XYZ classes, excess and no-movement inventory, safety stock and reorder points, plus buffer screening in the Advanced tier. The analysis is AI-assisted and reviewed and signed off by a CPIM- and DDMRP-certified planner. Ordered and paid through Upwork.
Data not ready yet? Get your ERP item master or inventory export cleaned and checked first (from $150): every change logged, nothing guessed.