Mean Absolute Deviation (MAD) Calculator

Mean Absolute Deviation (MAD) answers a simple question: on average, how far off is your forecast? It's the most direct, least mathematically dressed-up way to measure forecast accuracy — take every forecast error, ignore whether it was too high or too low, and average the size of the misses.

Forecasts are never perfect, and that's fine — the useful question isn't "was this forecast wrong," it's "how wrong, typically, and is that getting better or worse over time." MAD gives you a single number to track that, in the same units as the demand itself, which makes it intuitive: "our forecast is usually off by about 3.5 units" is immediately meaningful in a way a raw error percentage sometimes isn't.

For each period, subtract the forecast from the actual demand and take the absolute value — so an over-forecast and an under-forecast of the same size count equally, rather than canceling each other out. Average those absolute errors across all periods and that's MAD. A lower MAD means tighter, more accurate forecasts; there's no upper bound, since it scales with however large your errors happen to be.

Actual demand: 100, 120, 110, 130
Forecast: 105, 115, 108, 128

This forecast method is off by about 3.5 units per period, on average.

Compare a much rougher forecast on the same actual demand: 80, 140, 90, 150.

A MAD of 20 versus 3.5 makes the comparison obvious — the second forecasting method is missing by roughly six times as much, on average, even though both methods happened to average out to zero net bias across the period.

MAD treats every unit of error the same regardless of how big the underlying demand was, which makes it a poor tool for comparing accuracy across products with very different volumes — a MAD of 5 is excellent for a product that sells 1,000 units a week and alarming for one that sells 10. For that kind of cross-product comparison, MAPE (which expresses error as a percentage) usually makes more sense.

Comparing MAD across products of very different sizes.A MAD of 5 doesn't mean the same thing for a high-volume SKU and a low-volume one. Use MAPE instead when comparing across products.

Treating a single MAD reading as the full picture.MAD alone doesn't tell you whether errors are trending up, down, or holding steady — track it over time, not as a one-off snapshot.

Ignoring what MAD doesn't show — bias. Two forecasts can have the same MAD while one is consistently over-forecasting and the other under-forecasting. MAD hides that distinction since it only measures magnitude, not direction.

Never using it to actually compare methods. MAD is most useful side by side — moving average against exponential smoothing against whatever else you're considering. Calculating it for one method in isolation tells you less than comparing it across a few.

Chasing a lower MAD at any cost. An overly complex forecasting method that shaves a small amount off MAD may not be worth the added effort to maintain — weigh the improvement against the complexity.

There's no universal "good" MAD — it depends entirely on the scale of demand for the product being forecast. The meaningful benchmark is relative: track MAD as a percentage of average demand over time (which is effectively what MAPE does), or compare MAD across competing forecasting methods on the same product to see which one actually performs better.

Once you know how accurate your forecast is:

1. Check MAPE too — especially if comparing accuracy across products with different demand volumes.
2. Try adjusting your forecasting method — different alpha values or window sizes, then recheck MAD to see if it actually improved.

• Mean Absolute % ErrorA scale-independent version, useful for comparing across SKUs.
• Moving AverageTest whether this forecasting method is actually accurate.
• Exponential SmoothingTune alpha by checking which value produces the lowest MAD.

Books:

• Forecasting: Principles and Practice by Rob J Hyndman and George Athanasopoulos (free online)

Standards & curricula:

• APICS (ASCM) CSCP certification curriculum (demand planning module)

General references for further study, not endorsements — verify course availability and content directly with the provider.