Mean absolute deviation calculator forecast

Mean absolute deviation (MAD)
—

Enter demand and forecast values to get started.

Periods used
—
MSE
—
MAPE
—
Bias
—
RSFE
—
Tracking signal
—

Demand vs forecast

The shaded gap in each period is the forecast error. Orange gaps miss by more than the MAD.

demandforecasterror

Tracking signal

Running sum of errors ÷ running MAD, period by period. Inside ±4 the forecast is under control. Outside it, the forecast is biased.

Step-by-step working

How MAD and the tracking signal are calculated with your numbers.

    PeriodDemandForecastError|Error|RSFEMAD so farTS

    How mean absolute deviation calculator forecast works

    The mean absolute deviation calculator forecast works in 4 steps: it subtracts each period's forecast from actual demand, drops the minus signs, averages the absolute errors into the mean absolute deviation (MAD), and divides the running sum of forecast errors (RSFE) by MAD to get the tracking signal. MAD comes out in demand units, such as units, cases or kilograms (kg).

    1. EnterType demand and forecast per period, or fill the forecast with a naive, 3-period moving average or exponential smoothing method.
    2. Erroreₜ = Aₜ − Fₜ for every period.
    3. MADΣ|eₜ| ÷ n, the average miss per period.
    4. Tracking signalRSFE ÷ MAD, checked against ±4.

    The calculator reports 7 forecast accuracy measures: MAD, mean squared error (MSE), mean absolute percentage error (MAPE), bias, RSFE, tracking signal and the number of periods used. The example below covers 6 periods and gives MAD = 5.17 units.

    1201402P110P24P32P48P55P6
    Actual demand AₜForecast Fₜ
    The dashed gaps measure 2, 10, 4, 2, 8, 5 units. Their average, 5.17 units, is the forecast MAD.

    Mean absolute deviation calculator forecast formula

    The mean absolute deviation calculator forecast formula is MAD = Σ|Aₜ − Fₜ| ÷ n: the sum of absolute forecast errors divided by the number of periods.

    • MAD: mean absolute deviation of the forecast, the average miss per period.
    • Σ (sigma): add the absolute errors of all periods.
    • | |: absolute value. An over-forecast of 5 and an under-forecast of 5 both count as 5.
    • Aₜ: actual demand in period t.
    • Fₜ: forecast for period t.
    • n: the number of periods with a forecast.

    The formula with the example data

    MAD = (|120 − 118| + |135 − 125| + |128 − 132| + |140 − 138| + |150 − 142| + |145 − 150|) ÷ 6 = 31 ÷ 6 = 5.17

    6 forecast accuracy formulas

    MeasureFormulaExample value
    Mean absolute deviation (MAD)Σ|Aₜ − Fₜ| ÷ n5.17 units
    Mean squared error (MSE)Σ(Aₜ − Fₜ)² ÷ n35.5 units²
    Mean absolute percentage error (MAPE)Σ|(Aₜ − Fₜ) ÷ Aₜ| ÷ n × 100%3.73%
    Bias (mean error)Σ(Aₜ − Fₜ) ÷ n2.17 units
    Running sum of forecast errors (RSFE)Σ(Aₜ − Fₜ)13 units
    Tracking signalRSFE ÷ MAD2.52

    MAD converts to the standard deviation of forecast errors with σ ≈ 1.25 × MAD, for normally distributed errors. The example MAD of 5.17 units gives σ ≈ 6.46 units and a 95% safety stock of 1.65 × 6.46 ≈ 10.66 units.

    Mean absolute deviation calculator forecast technique

    The standard mean absolute deviation calculator forecast technique is the 8-column forecast error table: period, demand, forecast, error, absolute error, RSFE, running MAD and tracking signal, filled one period at a time. The running columns show the period where a forecast starts to drift.

    tAₜFₜeₜ|eₜ|RSFEMAD so farTracking signal
    112011822221
    213512510101262
    3128132-4485.331.5
    414013822104.52.22
    515014288185.23.46
    6145150-55135.172.52
    -5-2.502.551P12P21.5P32.22P43.46P52.52P6
    The shaded band runs from −4 to +4. Every period of the example stays inside the band, so the forecast shows no persistent bias.

    3 forecasting methods compared on periods 4 to 6

    0510155Given forecast9Naive (last period)11.223-period moving average
    The lowest MAD wins. Over periods 4 to 6, the given forecast misses by 5 units per period, the naive forecast by 9 and the 3-period moving average by 11.22.

    Compare methods on the same periods only. The naive method has no forecast for period 1 and the 3-period moving average has none for periods 1 to 3, so the comparison scores all 3 methods on periods 4 to 6.

    Frequently asked questions

    How do you calculate MAD in forecasting?

    Calculate MAD in forecasting by subtracting each forecast from actual demand, dropping the signs and averaging: MAD = Σ|Aₜ − Fₜ| ÷ n. For demand 120, 135, 128, 140, 150, 145 and forecasts 118, 125, 132, 138, 142, 150, the absolute errors are 2, 10, 4, 2, 8 and 5, the sum is 31 and MAD = 31 ÷ 6 = 5.17 units.

    What is a good tracking signal?

    A good tracking signal stays between −4 and +4. A tracking signal above +4 means the forecast runs low (demand beats the forecast period after period). A tracking signal below −4 means the forecast runs high. Some planners use a tighter ±3 limit for high-value items.

    How is forecast MAD used to set safety stock?

    Forecast MAD sets safety stock through the rule σ ≈ 1.25 × MAD: safety stock = z × 1.25 × MAD. For MAD = 5.17 units and a 95% service level (z = 1.65), safety stock = 1.65 × 1.25 × 5.17 ≈ 10.7 units. The 1.25 factor holds for normally distributed forecast errors.

    Is forecast MAD the same as mean absolute error?

    Yes, forecast MAD and mean absolute error (MAE) use the same calculation: the average of |actual − forecast|. Operations management and supply chain textbooks call the measure MAD (mean absolute deviation). Statistics and machine learning texts call the measure MAE.

    What does MAD mean in forecasting?

    MAD in forecasting means mean absolute deviation: the average size of the forecast error, Σ|demand − forecast| ÷ n. MAD uses the same calculation as mean absolute error (MAE). Operations and supply chain courses use the name MAD.

    What is the tracking signal?

    The tracking signal is the running sum of forecast errors (RSFE) divided by MAD. Errors that cancel out keep the tracking signal near 0. A forecast that keeps missing in 1 direction pushes the signal up or down, and many textbooks flag a forecast once the signal leaves ±4.

    How do I compare two forecasting methods?

    Compare 2 forecasting methods by running both on the same demand periods and picking the method with the lower MAD, or the lower MSE when large misses cost more. The fill buttons create a naive forecast, a 3-period moving average and an exponential smoothing forecast from your demand.

    What if some periods have no forecast?

    Type a dash (-) in the forecast list for any period with no forecast. The calculator leaves that period out of MAD and the other measures and keeps its demand on the chart.