# M3 Exponential Smoothing Forecast Calculator > Source: https://netray.co/tools/m3-holt-exponential-smoothing-forecast-calculator > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Free Tools ## Short answer Step a Holt level and trend forecast forward one period and see how the smoothing constant changes the result. ## Category infor-cloudsuite ## Erp Infor M3 ## Audience - M3 consultant - Demand planner - Supply chain analyst ## Tool **Kind:** calculator **Inputs:** #### Previous level **Key:** prev_level **Type:** number **Default:** 1000 **Min:** 0 **Step:** 10 **Unit:** units **Help:** The smoothed level from the last period. #### Previous trend **Key:** prev_trend **Type:** number **Default:** 12 **Step:** 1 **Unit:** units/period #### Latest actual demand **Key:** actual **Type:** number **Default:** 1060 **Min:** 0 **Step:** 10 **Unit:** units #### Level smoothing (alpha) **Key:** alpha **Type:** slider **Default:** 0.3 **Min:** 0.05 **Max:** 1 **Step:** 0.05 #### Trend smoothing (beta) **Key:** beta **Type:** slider **Default:** 0.1 **Min:** 0 **Max:** 1 **Step:** 0.05 #### Forecast horizon **Key:** horizon **Type:** number **Default:** 3 **Min:** 1 **Max:** 24 **Step:** 1 **Unit:** periods **Outputs:** #### Forecast that was made for this period **Key:** pred **Expr:** prev_level + prev_trend **Format:** number #### Forecast error (actual minus forecast) **Key:** err **Expr:** actual - pred **Format:** number #### Absolute error **Key:** err_pct **Expr:** abs(err) / actual * 100 **Format:** percent #### New level **Key:** level **Expr:** alpha * actual + (1 - alpha) * pred **Format:** number #### New trend **Key:** trend **Expr:** beta * (level - prev_level) + (1 - beta) * prev_trend **Format:** decimal #### Forecast for next period **Key:** fc_next **Expr:** level + trend **Format:** number **Primary:** true #### Forecast at horizon **Key:** fc_h **Expr:** level + horizon * trend **Format:** number **Primary:** true **Chart:** #### New level and next forecast vs alpha **Type:** line **X:** alpha **From:** 0.05 **To:** 1 **Steps:** 20 **Series:** - level - fc_next **Disclaimer:** Illustrates the standard Holt method one step at a time. Check how your M3 forecast model is actually parameterised before mapping these constants. ## Content **Intro:** Choosing smoothing constants by feel is how forecast models drift. This calculator walks one period of Holt's linear method so you can see exactly what alpha and beta do to level, trend and the next forecast. **Sections:** #### How to use it **Body:** Enter the level and trend carried forward from the previous period and the latest actual demand. Choose alpha for how fast the level reacts and beta for how fast the trend reacts. The calculator shows the forecast that was made for this period, the error, the updated level and trend, and the forecast for the next period and for a horizon of your choice. Repeat the step with the outputs as next period's starting values to roll the model forward. **Bullets:** - Higher alpha reacts faster but amplifies noise - Beta is usually kept small - Test the same period with two alpha values - Track error over many periods, not one #### The Holt method **Body:** The new level is alpha times the actual plus (1 minus alpha) times the old level plus old trend. The new trend is beta times the change in level plus (1 minus beta) times the old trend. The forecast h periods ahead is the level plus h times the trend. With beta set to zero the trend freezes, and with alpha at 1 the level simply equals the latest actual. #### Interpreting the output **Body:** A large absolute error with a stable trend usually means a one-off event, so avoid chasing it with a high alpha. Persistent error in one direction means the trend is not keeping up, so raise beta slightly. The chart shows how the new level moves from sluggish to fully reactive as alpha rises. Pick an alpha at which the level responds to genuine shifts but does not copy every spike. **Bullets:** - Typical alpha for stable items is 0.1 to 0.3 - Typical beta is 0.05 to 0.2 - Flag outliers before they enter the history #### Common mistakes **Body:** Applying a trend model to seasonal items produces forecasts that overshoot in the off season. Initialising level and trend badly also biases early periods. Another mistake is optimising alpha on a single month. Use a rolling error measure such as mean absolute deviation across at least a dozen periods, and compare against a naive forecast to confirm the model adds value. **Cta:** If you want forecast accuracy measured across all your M3 items and compared across models, Netray can build that analysis on your sales history. **Faq:** #### Does M3 use Holt exactly as shown? **Answer:** Do not assume so. Forecast models in any ERP differ in initialisation and parameter naming. Use this tool to understand behaviour and compare against your system's documented forecast model before mapping constants. #### What alpha should I start with? **Answer:** Start around 0.2 for stable demand and raise it only if error analysis shows the forecast lags real shifts. Values near 1 mostly echo the latest actual and give little smoothing. #### When should the trend be switched off? **Answer:** Set beta to 0 for flat or noisy items without a real direction. A trend fitted to noise extrapolates badly, particularly at long horizons. #### How do I handle seasonality? **Answer:** This tool does not include a seasonal component. For seasonal items, deseasonalise first or use a model with seasonal indices, then apply smoothing to the adjusted series. ## Search terms holt exponential smoothing calculator, m3 demand forecast, exponential smoothing alpha beta, trend forecast calculator, forecast smoothing constant, infor m3 forecasting, double exponential smoothing --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json