On the one hand, we can predict demand, because of the continuity of demand (for example, if the demand was large in the past, the demand will be large in the future), and on the other hand, because of the correlation of demand (for example, the greater the promotion, the greater the sales volume).
There are three basic forms of demand continuity: first, random changes, which means it goes up and down, but there is no trend or periodicity. Second, trend, which means that demand shows an increasing or decreasing trend over time. Third, cyclicality, that is to say, demand shows alternating peaks and troughs, and seasonality is one of them. Then there is the combination of trends and cycles.
For trending seasonal demand, this article introduces a commonly used forecasting method. The accuracy is not as high as the Holt-Winter model, but it is relatively simple and easy to implement, and everyone can implement it in an Excel sheet. Before introducing specific models, let’s first look at the difference between seasonality and cyclicality.
Cyclicity is a time series that exhibits wave-like ups and downs, usually caused by business and economic activities. It is different from a trend in that it is not a continuous movement in a single direction, but an alternating fluctuation of ups and downs; it is also different from seasonal changes, which have fixed rules, while cyclic fluctuations have no fixed rules (Baidu Encyclopedia, “Period” “Sex” entry). It can be said that both cyclicality and seasonality have peaks and troughs. The former lacks regularity and has low predictability; the latter has strong regularity and is easy to predict.
When people think of seasonality, they think of the four seasons throughout the year. That’s true, but seasonality doesn’t have to last a year: a day, a week, a month, even an hour or a minute can be seasonal.
For example, if you open a restaurant, the demand for breakfast, lunch and dinner is different. The need for labor is less in the morning and more at noon and evening. This is the seasonality of the day. Similarly, as a restaurant, Friday and Saturday nights are generally the busiest, while Tuesday nights are generally the busiest (this is why many restaurants in the United States have specials on Tuesdays), which is seasonal during the week.
E-commerce is similar, but the cycle is exactly opposite to that of restaurants: on weekends, little girls go out to go shopping and eat, and not many people shop online; on Tuesday, more people stay at home and don’t go out, and you will find that online shopping The volume will be higher – at least this is the situation in the United States. When I analyzed a cross-border e-commerce business, I saw a similar pattern.
If it is pure seasonality and there is no trend between quarters, we can calculate the ratio of each season to the average to predict the demand for each quarter of the next year.
To be as accurate as possible is to find a more appropriate forecasting method, and to improve the accuracy of the forecast as much as possible by optimizing each parameter; to correct the deviation as soon as possible is to have some actual sales data, analyze it as soon as possible, compare it with the original forecast, compare it with historical data, and then combine it with the market , sales, product management, etc., and adjust forecasts as soon as possible. By doing so, you can at least avoid big mistakes.
In addition, whether it is seasonality, cyclicality, or trend, it is relative to a specific supply chain response cycle and response capability.
When the response cycle of the supply chain is short and the response capacity is unlimited, the prediction of these variability is not so important. For example, if this is a front-end warehouse or a store, the replenishment cycle is only a few hours or a few days, which does not reflect much seasonality and trends, so you do not need to use complex models to predict – the moving average method. Simple exponential smoothing method, or even the “naive prediction” of selling one to make up for one, can handle it well in 80% of cases.