Root Mean Square Error (RMSE)

Root Mean Square Error (RMSE) measures how far forecasted values deviate from actual values. To calculate it, square each period’s forecast error, average the squared errors across all periods, then take the square root. RMSE quantifies how closely a forecast tracks actual data, and its magnitude is meaningful only relative to the mean of the time series.

RMSE at a Glance

The RMSE Formula and How to Calculate It

RMSE is calculated in four steps: find each period’s error, square it, average the squared errors, and take the square root. The formula is:

RMSE = √[ Σ (Fₜ − Aₜ)² ÷ n ]

where Fₜ is the forecast for period t, Aₜ is the actual value for period t, and n is the number of periods.

1. Compute each period’s error. Subtract the actual from the forecast (Fₜ − Aₜ). Positive errors mean you over-forecast; negative errors mean you under-forecast.
2. Square each error. Squaring removes the sign so over- and under-forecasts do not cancel out, and it gives larger errors more weight.
3. Average the squared errors. Sum the squared errors and divide by the number of periods. This result is the Mean Squared Error (MSE).
4. Take the square root. The square root of MSE returns the metric to the original units of the data. That value is RMSE.

Worked example: six periods of monthly demand

The table below applies the four steps to six months of demand for a single SKU.

Mean squared error = 13,400 ÷ 6 = 2,233.3. RMSE = √2,233.3 = 47.3 units. Against mean demand of 1,020 units, an RMSE of 47.3 means the forecast typically misses by about 4.6% of average demand.

How to Interpret RMSE

There is no universal “good” RMSE value. You evaluate RMSE by comparing it to the mean of the time series it measures. Lower is always better, and because RMSE is expressed in the same units as the data, planners can read it directly as “the forecast typically misses by about this many units.”

Scale is what gives RMSE its meaning. An RMSE of 50 units is a tight forecast for an item with mean demand of 1,000 units (5% of the mean). The same RMSE of 50 is a poor forecast for an item with mean demand of 100 units (50% of the mean). Dividing RMSE by the series mean produces a normalized RMSE (sometimes called the coefficient of variation of RMSE), which makes this comparison explicit.

RMSE vs. MAE vs. MAPE

RMSE, MAE, and MAPE all measure forecast error, but they weight errors differently. RMSE penalizes large errors more heavily because it squares them. MAE treats every error equally. MAPE expresses error as a percentage, which makes it scale-free but unreliable when actual values are near zero.

 
In the six-month example above, RMSE is 47.3 units, MAE is 43.3 units, and MAPE is 4.2%. RMSE sits above MAE because the 70-unit miss in month 3 carries extra weight once squared. RMSE is always greater than or equal to MAE for the same data; the wider the gap, the more your error is concentrated in a few large misses.

Why RMSE Matters in Demand Forecasting

RMSE matters in demand forecasting because it connects forecast error directly to inventory risk. Planners use it for three jobs: comparing forecasting methods, monitoring forecast performance over time, and sizing safety stock.

Comparing forecasting methods on the same series

When you backtest two or more statistical models against the same history, the model with the lower RMSE produced smaller errors, with extra credit for avoiding big misses. Because every model is scored on the same series, scale dependence is not an issue. Statistical forecasting methods can be ranked this way before one is promoted to production.

Monitoring forecast performance over time

Tracking RMSE on a rolling window (for example, the last 6 or 12 periods) shows whether a forecast is degrading. A rising RMSE is an early signal of a demand shift, a new customer pattern, or a model that no longer fits. Pair it with a bias measure, because RMSE shows the size of error but not its direction. See detecting forecast bias early for how Tracking Signal complements RMSE.

Setting safety stock

RMSE is a practical input for safety stock because it estimates the variability of forecast error, which is the uncertainty safety stock exists to cover. A common formulation is:

Safety stock = z × RMSE × √(lead time in periods)

where z is the service-level factor (1.65 for roughly a 95% cycle service level). Using the example above, with RMSE of 47.3 units and a four-month lead time: 1.65 × 47.3 × √4 ≈ 156 units of safety stock. This approach assumes the forecast is roughly unbiased; the example’s average error is −6.7 units, small relative to RMSE. When forecast error shrinks, safety stock can shrink with it. That is why forecast accuracy and safety stock calculation belong in the same conversation, and why Arkieva Inventory Planning ties buffer levels to measured forecast error rather than rules of thumb.

Limitations of RMSE

RMSE has three main limitations: it is sensitive to outliers, it depends on scale, and it says nothing about bias on its own. None of these make RMSE a bad metric; they mean it should not be the only metric.

• Outlier sensitivity. One unusually large error can inflate RMSE far more than MAE. In the example above, if month 3’s miss grew from 70 to 170 units, RMSE would rise 67% (47.3 → 79.0) while MAE would rise 38% (43.3 → 60.0). Decide whether that sensitivity is what you want before relying on it.
• Scale dependence. Because RMSE is in the data’s units, you cannot compare RMSE across items with different volumes. Use MAPE, a weighted MAPE, or normalized RMSE (RMSE ÷ mean) for portfolio-level comparisons.
• No direction. RMSE treats an over-forecast and an under-forecast of the same size identically. A forecast that is consistently high can post a respectable RMSE while steadily building excess inventory.

The practical answer is to track a small set of metrics together: RMSE for the cost of large misses, MAE or MAPE for typical error, and a bias measure for direction. Hyndman and Athanasopoulos make the same recommendation in Forecasting: Principles and Practice (3rd ed., 2021), noting that scale-dependent measures such as RMSE and MAE should only be compared on the same series. For more on building a balanced scorecard, see forecast accuracy KPIs.

Frequently Asked Questions

Put Forecast Error to Work

Measuring RMSE is the first step. Using it to choose better models, catch forecast drift early, and set safety stock that reflects real uncertainty is where the value is. Arkieva Demand Planning calculates and tracks forecast error metrics across every item and level of your hierarchy, so planners spend their time acting on the numbers instead of assembling them.

Ask for a one-on-one demo of the Arkieva supply chain planning solution → Request a Demo