Exponential Smoothing Forecast Calculator

Inputs

• Periods of history you are entering — Between 3 and 12. The demand rows below follow this number, so eight quarters of history is eight boxes, not twelve with four to ignore.
• Period 1 demand — The oldest period. Months, weeks, or quarters, as long as every row is the same length.
• Period 2 demand
• Period 3 demand
• Period 4 demand
• Period 5 demand
• Period 6 demand
• Period 7 demand
• Period 8 demand
• Period 9 demand
• Period 10 demand
• Period 11 demand
• Period 12 demand
• Smoothing constant (alpha) — Between 0 and 1, as a decimal: 0.3, not 30%. It is the share of this period's surprise that moves the forecast. Low is steady and slow to react, high chases every wobble.
• Trend constant (beta) — Leave it at 0 for simple exponential smoothing, which forecasts a flat line. Raise it and the forecast learns a slope, which is what a climbing or falling series needs.

How to use this calculator

1. Enter how many periods of history you have, then type the demand for each one, oldest first. Months, weeks or quarters all work, as long as every row is the same length of time. Leave the rest at their example figures; anything past your count is ignored.
2. Set the smoothing constant α. It is the share of each period's surprise that moves the forecast: 0.1 is slow and steady, 0.5 reacts fast and is easily fooled by one odd week. Most demand series settle somewhere between 0.1 and 0.3.
3. Leave the trend constant β at 0 for simple exponential smoothing. If your demand is climbing or falling steadily, raise β and watch the error figures: a level-only model cannot forecast a slope, and it will lag one forever.
4. Read the tracking signal before anything else. Inside ±4 the forecast is missing in both directions, which is ordinary. Outside it, the forecast is wrong the same way every period, and that is a problem no smoothing constant fixes.
5. Check the two rival forecasts underneath. If "next period equals this one" beats your model on the same periods, the model is not earning its place, and the honest move is a simpler rule or a longer look at what drives the demand.
6. Carry the MAD or the RMSE into the safety stock calculation. Safety stock exists to cover what you did not forecast, so the spread of your forecast ERROR is the right input, not the spread of demand itself.

What each term means

Smoothing constant (alpha)

How much of the latest period's surprise goes into the next forecast. At 0.3, each new observation carries 30% and the memory of any single period halves in about two periods.

Trend constant (beta)

The same idea applied to the SLOPE: how fast the model updates its view of how quickly demand is rising or falling. At 0, there is no slope and the forecast is a flat line forever.

MAD (mean absolute deviation)

The average size of a miss, ignoring direction. In the same units as demand, which is what makes it the usual input to a safety stock calculation.

RMSE (root mean squared error)

The same average miss, but squared before averaging, so one bad period weighs far more than several small ones. Use it when a large miss hurts more than proportionally.

MAPE

The average miss as a percentage of actual demand, which makes two products of different sizes comparable. It is undefined when demand is zero and exaggerates the error on very small periods.

Tracking signal

Cumulative error divided by MAD: how many average misses the forecast has drifted, in one direction, over the whole run. Past ±4 it is conventionally called out of control.

Naive forecast

"Next period equals this one." It costs nothing, it is surprisingly hard to beat on short horizons, and it is the bar any real forecasting method has to clear to justify itself.

More Forecasting tools

• Seasonal Index & Forecast Calculator
• Forecast Accuracy & Tracking Signal Calculator