Forecasting
Forecast Accuracy & Tracking Signal Calculator
Grade a forecast you already have. Put your forecast for each period beside what actually happened and this returns the four error measures an operations course asks for — MAD, MSE, RMSE and MAPE — plus the one that matters most and gets quoted least: the tracking signal, which separates a forecast that is merely imprecise from one that is wrong in the same direction every single period.
Inputs
- Periods you are entering — Between 3 and 8. The rows below follow this number, oldest first.
- Period 1 — actual
- Period 1 — forecast — What you predicted for that period BEFORE it happened, whatever produced it: a model, a sales plan, a guess.
- Period 2 — actual
- Period 2 — forecast
- Period 3 — actual
- Period 3 — forecast
- Period 4 — actual
- Period 4 — forecast
- Period 5 — actual
- Period 5 — forecast
- Period 6 — actual
- Period 6 — forecast
- Period 7 — actual
- Period 7 — forecast
- Period 8 — actual
- Period 8 — forecast
How to use this calculator
- For each period, enter what actually happened and what you had forecast for it BEFORE it happened. The forecast can come from anywhere: a model, a sales plan, a supplier's guess.
- Read the tracking signal first. It is the one number that separates a forecast that is merely imprecise from one that is wrong in the same direction every period, and only the second kind can be fixed by changing the method.
- Use MAD when a miss of 200 units is twice as bad as a miss of 100, and RMSE when it is four times as bad, which is the case whenever a big miss means expedited freight or a stopped line.
- Compare MAD with the naive rule beneath it. If assuming next period equals this one would have done better, whatever produced this forecast is adding noise rather than information, however sophisticated it is.
- Look at the largest single miss before you act on the averages. One bad period can carry MAD on its own, and if that period had a cause you can name, the right fix is to model the cause rather than retune the forecast.
- Take MAD to the safety stock calculation as the variability you have to buffer, and take the bias to whoever owns the forecast. A lean of 30 units a period is 360 units a year of stock you either never ordered or never needed.
What each term means
- Error
- Actual minus forecast. Positive means demand beat the forecast and you were short; negative means you forecast more than arrived. Some courses define it the other way and flip every sign.
- Bias (mean error)
- The average error WITH its sign, so over- and under-forecasts cancel. Near zero is what you want. Far from zero means the forecast is systematically off, which costs far more than the same error scattered randomly.
- RSFE
- Running sum of forecast errors: every period's error added up, signs included. It is the cumulative units you were short or long across the whole run.
- Tracking signal
- RSFE divided by MAD. It counts how many average misses the forecast has drifted in one direction. Outside ±4, investigate the method rather than the constants.
- MASE
- MAD divided by the MAD of the naive rule on the same data. Below 1 the forecast beats doing nothing; at or above 1 it does not. It is the only figure here that survives a change of units or product.
- MSE and RMSE
- Mean squared error and its square root. Squaring punishes large misses disproportionately, so a method chosen on RMSE will be more cautious about the occasional disaster than one chosen on MAD.
Educational disclaimer: Outputs are simplified educational estimates built from the numbers you enter — they are not financial, legal, tax, or investment advice, and real decisions deserve verified figures and qualified professionals.