Scroll Top
100 Powdermill Road, Suite 108, Acton, MA 01720

Demand Planning 101: Forecasting with Seasonality and Cyclicality

INTRODUCTION: FORECASTING WITH SEASONALITY AND CYCLICALITY

Whenever I’m involved in demand planning projects at client companies, there’s always intense interest — and concern — among stakeholders about how the forecasting algorithms work, particularly among those new to supply chain planning and demand forecasting. The forecasting process is viewed as a black box that produces magical numbers that the organization will march to, but no one really understands what’s inside the box. The fact that there’s a lot of mathematical jargon associated with forecasting adds to its mystique. So when I decided to launch a series of blogs on the basics of demand planning, I thought I’d begin with a discussion of forecasting. And a good place to start is to discuss forecasting with seasonality and cyclicality.

Demand forecasting involves analyzing past demand, determining the different factors driving demand, and using that knowledge to predict future demand. For many products, a large portion of the variation in demand over time can be attributed to seasonality and other cyclical patterns. In this article, we’ll explore the basics of seasonality and cyclicality, and discuss some techniques for addressing such patterns. These techniques describe how a forecaster would manually create a forecast and how a software application (such as New Horizon Demand Planning) would do the same.

Understanding Seasonality and Cyclicality

Seasonality refers to predictable patterns of demand that fluctuate based on annual seasons. For example, a retailer that sells winter coats may experience a spike in demand during the winter months, and lower demand during the summer months. Seasonality can be caused by a variety of factors, including weather patterns, holidays, and cultural or social events.

Cyclicality, on the other hand, refers to a broader set of periodic fluctuations in demand that occur in a predictable pattern not limited to seasonal patterns. For example, a company that sells construction materials may experience higher demand during certain phases of the economic cycle, such as when housing starts are on the rise. Macroeconomic factors are a major source of cyclicality, but there can also be other causes.

Both seasonality and cyclicality can present challenges for supply chain demand planners, as they require accurate forecasting of demand patterns over time. However, several techniques can help planners account for these factors and improve the accuracy of their forecasts.

Detecting Seasonality and Cyclicality

The first step in accounting for seasonality and cyclicality in demand planning is to detect the patterns in your historical sales data. There are several techniques you can use to do this, including:

  • Visual inspection: One of the simplest ways to detect seasonality and cyclicality is to plot your sales data over time and look for recurring patterns. You may notice spikes in demand around certain holidays or events, or regular fluctuations that occur over the course of the year or the economic cycle.
  • Seasonal decomposition: Seasonal decomposition is a technique for breaking down a time series into its component parts – trend, seasonal, and residual – using mathematical models such as the classical decomposition or the X-11 method. Seasonal components can then be isolated and analyzed to identify patterns and effects related to seasonality and annual holidays.
  • Autocorrelation analysis: Autocorrelation analysis involves examining the autocorrelation function (ACF) of a time series to detect any patterns related to seasonality or annual holidays. A significant spike in the ACF at a specific lag indicates a strong correlation between values separated by that lag, which may be related to seasonal or holiday effects.
  • Box-Jenkins models: Box-Jenkins models involve fitting an autoregressive integrated moving average (ARIMA) model to a time series, which includes terms for autoregression, moving averages, and differencing. Seasonal ARIMA models (SARIMA) extend this approach to incorporate seasonal components. These models can detect and account for seasonality and holiday effects in the data.

Using Forecasting Techniques to Account for Seasonality and Cyclicality

Once you have detected the patterns of seasonality and cyclicality in your historical sales data, you can use this information to develop forecasts that account for these effects. Some techniques you may want to consider include:

  • Seasonal forecasting: Seasonal forecasting involves developing separate forecasts for each season or holiday period, based on historical sales data for that specific time period. For example, if you are forecasting demand for winter coats, you may develop a separate forecast for each winter season based on past sales data for that time
  • Time series decomposition: Time series decomposition involves breaking a time series down into its component parts, including trend, seasonality, and residual components. This allows you to develop forecasts for each component separately, and then combine them to create an overall forecast. This approach can be especially useful when multiple seasonal or cyclical patterns are present in the data.
  • Regression analysis: Regression analysis involves using historical sales data along with additional variables, such as economic indicators, to develop a model that can predict future demand. This approach can be useful when there are factors beyond seasonality and cyclicality that are influencing demand. (This technique is aimed at accounting for factors other than seasonality and cyclicality, but it may be necessary to do this to correctly account for seasonal and cyclical effects.)
  • Exponential smoothing: Exponential smoothing is a forecasting method that assigns greater weight to recent data points and less weight to older data points, with the goal of capturing trends and patterns in the data. This approach can be adapted to account for seasonality and cyclicality, using methods such as seasonal exponential smoothing or Holt-Winters exponential smoothing.

Improving the Accuracy of Your Forecasts

Regardless of the forecasting technique you choose, there are several steps you can take to improve the accuracy of your forecasts:

  • Use multiple methods: Try using multiple forecasting methods to generate a range of potential outcomes. This can help you identify trends and patterns that may not be captured by a single method.
  • Use up-to-date data: Make sure that your historical sales data is up-to-date and relevant to current market conditions. This can help you account for changes in consumer behavior or economic trends that may affect demand.
  • Consider external factors: Take into account external factors that may influence demand, such as changes in weather patterns, shifts in consumer preferences, or disruptive events such as natural disasters.
  • Continuously monitor and adjust your forecasts: Continuously monitor your forecasts and adjust them as needed based on new information. This can help you respond to changing market conditions and ensure you have the right amount of inventory to meet customer demand.

Conclusion

For many products, accounting for seasonality and cyclicality is an essential first step toward developing an accurate forecast. Once you have an understanding of such effects, you can more accurately analyze and account for other drivers of demand to come up with the most accurate forecast possible.

To Learn More

To learn how New Horizon can help you with your demand planning challenges, contact us – we’d love to talk.