The role of seasonality in demand forecasting success

Detecting seasonality

You probably have a general sense of your product’s seasonality, but turning it into useful input for forecast requires data. Specifically, store point-of-sale (POS) and e-commerce sales. Customer orders are not a good reflection of true consumer demand, nor for detecting seasonality, because retailers muddy the waters with their ordering algorithms and decisions to stock up or wind down inventory for different reasons.

To check if a specific time series exhibits seasonality, you can perform hypothesis testing on it. Another option is to take a more empirical approach and check whether models with or without a seasonal factor are more representative of your data using a measure like the Akaike information criterion.

Seasonality will typically vary from region to region, depending on local calendars and weather. You may need to segment your data geographically to identify seasonality patterns. At the same time, aggregating forecasts to a less granular level—product category instead of product, for example—may make it easier to distinguish seasonal patterns from random noise.

Modeling methodologies

Most forecasting methodologies allow for explicit modeling of a seasonal term. SARIMA (ARIMA with seasonality) allows for forecasts based solely on the past values of the forecast variable. The Holt-Winters seasonal method comprises a forecast equation and three smoothing components for the level, trend and seasonal components.

But beware of their limitations and additional accommodations you may need to make depending on the model you select. Some models don’t allow for multiple seasonality, like both weekly and yearly. If you’re using one of these, you have a few options depending on the nature of your data:

• Choose to include only the most critical type of seasonality
• Go with seasonality as a categorical variable
• Model the cyclic nature of seasonality by using Fourier terms as regressors

Then there are some models that assume integer seasonality. That’s problematic when there are 365.25 days in a 12-month period due to leap years, and the length of a month fluctuates throughout the year.

What’s more, some seasonal events aren’t tied to a calendar date (Lunar New Year, Thanksgiving, etc.), so a seasonal term isn’t the best way to account for them. Instead, you could model them as binary regressors using holiday calendars.

Meeting the seasonality challenge

Once you have a solid baseline forecast, your seasonality picture will become much clearer, so it’s worth the investment to get it right. A good rule of thumb: Go back as many years as possible in order to pinpoint reliable seasonality patterns. When it comes to newer products or product categories, you may need to extract those patterns from similar products or by leveraging syndicated market data.

Then you can focus on the next step for a best-in-class demand forecasting process, making the appropriate forecast adjustments to account for coming events and product dynamics. It’s where the most variability can occur, and so where your demand planners should spend more of their time.