TradeGecko's Demand Forecasting uses Smooth Moving Average model. This model uses a series of averages over a long look back period. Spikes in sales are smoothed out to give you the prevailing trends.
What are the benefits of using this model?
- This model works great for predicting sales trends of products that are stable.
- When there is one day of high sales during the week, it is averaged out over the week.
- This model works well on products that are growing consistently and products that are declining over time.
To see how the forecasted number was calculated, click on the "chart" icon found on the variant forecast detail. It will show you the sales trends that were used to generate your forecast.
Understanding how this model works
We start with a look back period of 2 years and start forecasting from the first day you had sales within that 2-year window. You can select week or month as your aggregation period. This determines the window we use to average your sales. If you select 'Week', the window we use will be 7 days.
Predicting for 1 day
Let's assume that you only started selling this variant on the 1st of Jan. After 7 days, you have sold 70 units. On the 8th of Jan, you create a forecast on TradeGecko. We calculate the average to be 10 units and predict that you will sell 10 units on the 8th of Jan.
Predicting for 2 days
To predict for the 9th of Jan, we take the old total, subtract the previous average, add the new data point, and divide it by the same window of 7. The new average is still 10. This is because the window has not moved yet.
Predicting for 3 days
To predict for the 10th of Jan, we take the old total again. This time, the window has shifted. The total is now 80. We subtract the previous average, add the new data point, and divide it by the same window of 7. The new average is 11.14. Rounding to the nearest whole number only takes place after we sum all forecasts for the week.
Predicting for N days
Assuming that you want to build a forecast for 2 weeks. We will generate 14 new averages and then sum the total. This gives you the number of forecasted sales for the 2 weeks. Your recommended reorder quantity is the total incoming stock subtracted from the forecasted quantity.
Minimum days of sales
SKUs with no sales data in the last 2 years will not appear in the forecast. Furthermore, we require a product to have at least 50 days of sales to generate a forecast.
SKUs with less than 50 days of sales history will still appear in the forecast but will not have a stock out date and forecast quantity.