How To Calculate Mean Forecast Error

Use this premium calculator to measure average forecast bias from your actual and predicted values. Mean Forecast Error, often abbreviated as MFE, helps reveal whether a forecasting method tends to overestimate or underestimate outcomes over time.

Paste your actual and forecast numbers, choose your error direction, and instantly calculate the sum of errors, mean forecast error, and a visual error chart for fast interpretation.

Mean Forecast Error Calculator

Enter matching series with commas or line breaks. Example actual values: 120, 135, 128, 142. Example forecasts: 118, 140, 130, 145.

Expert Guide: How to Calculate Mean Forecast Error

Mean Forecast Error is one of the most useful statistics for identifying directional bias in forecasting. If you are trying to understand whether your model usually predicts too high or too low, MFE gives you a direct answer. Unlike magnitude-based metrics such as Mean Absolute Error or Root Mean Squared Error, Mean Forecast Error preserves the sign of each error. That means positive and negative forecast errors do not disappear when you average them. Instead, they show the net tendency of the forecast process.

In the simplest terms, Mean Forecast Error equals the arithmetic mean of the forecast errors across a set of observations. The exact sign depends on which convention you use. Many analysts define error as Actual minus Forecast. Under that convention, a positive MFE means the forecast was, on average, too low. Others define error as Forecast minus Actual, which reverses the interpretation. Because both are common in business analytics, supply chain planning, energy demand modeling, and academic forecasting research, the most important practice is consistency. Your reports should clearly state which error formula you used.

Core formula: MFE = Sum of forecast errors / Number of forecasts. If error = Actual – Forecast, then MFE = Σ(A – F) / n.

Why Mean Forecast Error matters

Forecasting is not only about being close to the truth. It is also about avoiding systematic bias. A model may look accurate in some periods but still regularly lean high or low. This matters in real operations. In retail and inventory management, persistent underforecasting can lead to stockouts, emergency replenishment costs, and lost sales. In budgeting, overforecasting revenue can create unrealistic expectations and poor capital allocation. In public planning, biased forecasts can distort staffing, resource planning, and service levels.

• MFE identifies direction: It tells you whether forecasts are generally above or below actual outcomes.
• MFE supports model monitoring: A drift in MFE over time can indicate changing market behavior or a stale forecasting model.
• MFE improves accountability: Teams can see whether bias comes from assumptions, data issues, seasonality, or human overrides.
• MFE complements other metrics: It should be used alongside MAE, MAPE, RMSE, and tracking signals for a full view.

Step by step: how to calculate mean forecast error

The process is straightforward when your actual and forecast values are paired in the same order. Each actual observation must match the forecast created for that specific period or item. Once your data is aligned, calculate the difference for every pair, sum those differences, and divide by the total number of observations.

1. List the actual values for each period.
2. List the corresponding forecast values for the same periods.
3. Choose your formula convention: Actual – Forecast or Forecast – Actual.
4. Calculate the error for each observation.
5. Add all individual errors together.
6. Divide the total error by the number of observations.
7. Interpret the sign to determine overall bias.

For example, suppose actual monthly demand is 120, 135, 128, and 142. The forecast values are 118, 140, 130, and 145. If we define error as Actual minus Forecast, the individual errors are 2, -5, -2, and -3. The sum of those errors is -8. Divide -8 by 4 and the MFE is -2. This indicates the forecast was, on average, 2 units too high. If you instead defined error as Forecast minus Actual, the MFE would be +2, which communicates the same bias but with the opposite sign convention.

Mean forecast error formula details

The general formula can be written as:

MFE = (e₁ + e₂ + … + e_n) / n

Where each error term e_t is either A_t – F_t or F_t – A_t, depending on the convention. The value of n is the total number of paired observations.

One of the most important things to understand is that MFE can hide large offsetting errors. For example, if one period has a +20 error and another has a -20 error, the average is zero. That does not mean the forecasts were perfect. It only means there was no net directional bias across those periods. For this reason, MFE should never be the only metric you review.

Interpreting positive, negative, and zero MFE

The interpretation depends on your sign convention, so always state the formula in your report. Using the common Actual – Forecast convention:

• Positive MFE: Forecasts were too low on average. Actual outcomes exceeded forecast values.
• Negative MFE: Forecasts were too high on average. Forecast values exceeded actual outcomes.
• MFE near zero: There is little net bias, but large positive and negative errors may still exist.

Metric	Formula	What it Measures	Best Use Case
Mean Forecast Error	Σ(A – F) / n	Directional bias	Checking whether a forecast is systematically high or low
Mean Absolute Error	Σ|A – F| / n	Average error magnitude	Measuring average size of misses without cancellation
Mean Absolute Percentage Error	Σ(|A – F| / A) × 100 / n	Relative percentage error	Comparing performance across products or scales
Root Mean Squared Error	√[Σ(A – F)² / n]	Error magnitude with heavier penalty on large misses	When large errors are especially costly

Worked example with realistic business data

Imagine a warehouse forecasting weekly shipments over eight weeks. Actual shipments are 1010, 980, 1045, 990, 1035, 1005, 1020, and 995. Forecasts are 1000, 1005, 1030, 1000, 1040, 1010, 1035, and 1005. Using Actual – Forecast, the errors are 10, -25, 15, -10, -5, -5, -15, and -10. The total is -45. Divide by 8 and the MFE is -5.625 shipments. This means the forecast is biased upward by about 5.63 shipments on average.

In a small series, that may not seem large. But in operations with thousands of SKUs and many periods, even modest bias can have measurable cost impacts. A persistent MFE of only a few units per week may compound into excess inventory, avoidable markdowns, or service failures when spread across an annual planning horizon. That is why planners often track MFE at multiple levels, such as item, category, region, and total portfolio.

Real-world statistics and benchmark context

Forecast performance varies sharply by industry, horizon, and volatility. Publicly available benchmark sources show that forecast quality often degrades as time horizons lengthen or when systems face structural shocks. For example, macroeconomic and energy forecasts are inherently uncertain because they respond to policy changes, weather variation, commodity prices, and behavioral shifts. In retail and demand planning, promotional lift, seasonality, and stockout distortions can all bias the data generating process.

Forecasting Context	Illustrative Error Pattern	Common Bias Risk	Operational Meaning
Short-term retail demand	MAPE often in the 10% to 30% range for stable items, higher for intermittent demand	Promotional overforecasting or stockout-driven underforecasting	Bias affects inventory turns and service levels
Energy load forecasting	Day-ahead system forecasts can achieve low single-digit percentage errors in mature markets	Weather model shifts can create directional bias	Bias affects procurement and balancing costs
Macroeconomic projections	Forecast revisions can be large during turning points and recessions	Persistent optimism or pessimism in assumptions	Bias affects budgets, rates, and public planning
Hospital utilization planning	Census and staffing forecasts can swing materially during outbreaks or seasonal surges	Systematic underforecasting during demand spikes	Bias affects staffing, beds, and emergency response capacity

These ranges are context-dependent and should be treated as directional examples rather than universal standards. What matters for MFE is whether the average signed error remains close to zero over a relevant review period. A company may tolerate some random error, but persistent positive or negative MFE indicates systematic bias that should be investigated.

Common mistakes when calculating MFE

• Mismatching periods: If the forecast for March is accidentally compared with actual April results, the metric becomes meaningless.
• Mixing conventions: Switching between Actual – Forecast and Forecast – Actual without disclosure will reverse the sign.
• Ignoring missing values: Every actual value should have a corresponding forecast. Incomplete pairing distorts the average.
• Using MFE alone: A zero MFE can occur even when absolute errors are large.
• Combining incompatible groups: Aggregating very different products or regions can hide localized bias.

How to use MFE to improve forecasts

MFE becomes most valuable when it is part of a feedback loop. If your MFE is consistently positive under the Actual – Forecast convention, your model may be too conservative. You may need to adjust trend assumptions, update seasonality factors, or review whether demand has structurally shifted upward. If MFE is consistently negative, the model may be overly optimistic. That can happen when recent surges are projected too far forward, promotional assumptions are inflated, or the baseline model is not adapting to softening demand.

1. Track MFE by time period and by product or segment.
2. Set review thresholds for acceptable bias.
3. Investigate process changes when MFE drifts materially.
4. Pair MFE with MAE or RMSE to distinguish bias from volatility.
5. Document whether bias comes from the statistical model or manual overrides.

MFE versus tracking signal

Many planners use MFE alongside a tracking signal. The tracking signal compares cumulative forecast error to a scaled measure of average absolute error. It helps determine whether bias is not only present but large relative to normal forecast variability. In practical terms, MFE tells you the average signed error, while the tracking signal helps determine whether that bias is becoming operationally significant.

When MFE is especially useful

MFE is especially useful in recurring forecasting environments where directional bias has a real cost. Supply chain planning, labor scheduling, energy dispatch, public budget projections, hospital occupancy planning, and recurring subscription revenue forecasting are all examples where a model that is consistently high or low creates operational friction. If your stakeholders ask, “Are we generally overforecasting or underforecasting?” MFE is one of the best first metrics to show.

Authoritative sources for deeper study

For additional reading on forecasting principles, error analysis, and data practices, review these authoritative resources:

• U.S. Census Bureau: Time Series Analysis resources
• U.S. Energy Information Administration: Forecast and analysis publications
• Monash University: Forecasting Principles and Practice

Final takeaway

Mean Forecast Error is simple to compute, easy to explain, and extremely useful for diagnosing forecast bias. The calculation is just the average of signed forecast errors, but the interpretation can reveal whether your forecasting process systematically leans high or low. That makes MFE a practical management metric, not just a technical statistic. Use it consistently, report the sign convention clearly, and always pair it with a magnitude-based measure so you do not overlook large offsetting misses. If your goal is to build reliable forecasts and improve planning discipline, MFE deserves a permanent place in your dashboard.

Note: Forecast error standards differ by industry, data frequency, product mix, and planning horizon. The most useful benchmark is usually your own historical performance over time, segmented in a way that reflects the decisions your organization actually makes.