Forecasting

Forecast Accuracy & Bias Analyzer

Measure how close a forecast came, and whether the misses cancel out or all lean the same way. A forecast that is wrong at random is a different problem from one that is always high.

Input data

One row per period. Fill Item to also see accuracy by item.

Number format
Forecast against actual by period.
PeriodItemForecastActual

Tip: paste a rectangular range from Excel or Sheets.

Your financial data is processed locally in your browser and is not uploaded to Consilium.

Forecast accuracy

92.3%

The forecast for revenue was 92.3% accurate across 6 periods. On average it over-forecast by $65.0 a period. 76.5% of the total error is one-directional, and the forecast was above actual in 5 of 6 periods.

WAPE
7.7%
MAPE
7.9%
Mean error
-$65.0
Bias share
76.5%
The mean error is -$65.0 a period, which is an over-forecast. Forecast was above actual in 5 of 6 periods, below in 1. Tracking signal -4.58824.

Error by period

Actual less forecast. Bars above the line are under-forecasts; below, over-forecasts.

Reconciled
PeriodForecastActualErrorPercentDirection
M1$1,000.0$900.0-$100.0-11.1%Over-forecast
M2$1,100.0$1,000.0-$100.0-10.0%Over-forecast
M3$1,200.0$1,150.0-$50.0-4.3%Over-forecast
M4$1,000.0$1,060.0+$60.05.7%Under-forecast
M5$1,300.0$1,200.0-$100.0-8.3%Over-forecast
M6$1,400.0$1,300.0-$100.0-7.7%Over-forecast
Total$7,000.0$6,610.0-$390.0

Reuse the analysis

Copy the summary, or export the periods and the chart.

Transparent by design

How this calculation works

There is one sign convention and it is named everywhere: error is actual less forecast. A negative error means the forecast was above the actual. It is stated in the column heading, here, and in the CSV header.

Error

Actual less forecast, every time. Negative is an over-forecast, positive an under-forecast, and both are also named in words so the sign is never the only signal.

WAPE and accuracy

Total absolute error divided by total absolute actual, then subtracted from one. Weighting by size stops a tiny period with a large percentage miss dominating the answer.

MAPE

The plain mean of the percentage misses. Periods where a percentage cannot be struck are excluded from this figure only — their currency error still counts in everything else.

Mean error, the bias

The average signed error. Where the misses cancel out it is near zero even when each period was badly wrong, which is why it is shown beside the absolute measures and never instead of them.

Bias share

The part of the total miss that did not cancel out. A high share means the error is one direction repeated rather than noise, which is a different problem with a different fix.

Tracking signal

Summed error divided by mean absolute deviation. A conventional check on whether a forecast has drifted persistently in one direction.

Suppressed percentages

An actual of zero has no base to divide by. A forecast and actual on opposite sides of zero would produce a percentage that describes nothing. Both are listed with the reason rather than shown as a number.

Reconciliation

Total forecast plus the summed error equals total actual. Any residual is shown rather than absorbed.