Advanced Forecasting

What is forecasting?

Nowadays, all companies create forecasts for the upcoming days/months/quarters/years. This can be done for procurement, sales but also other financial variables. These forecasts are based on business knowledge and statistical forecasting methods; most often, traditional statistical forecasting methods. Forecasting is important for business, because companies base important decisions on them.

Why advanced forecasting?

Over the last few years, there has been a lot of development in the world of statistical forecasting. In fact, with the Fourth Makridakis Competition in 2020, it was finally proven that complex forecasting methods, such as those leveraging machine learning and specifically neural networks, are better than the simpler traditional models. Three of the newer approaches to forecasting include:

• Machine learning – the regression approach
• (Recurrent) Neural Network
• Hybrid model (combination between RNN and the traditional forecasting methods)

While these typically use past (historical) data to make forecasts, we can try to incorporate external features into the forecasting algorithm if we wish to further improve the accuracy.

Example

For example, Jonathan sells ice cream. He has two variables in his dataset, the amount of ice cream sold per day (1) and the matching date (2). Based on these two variables, Jonathan can create a forecast for the upcoming year. The forecast might be useful, but could always be improved, which might give Jonathan more information to help make better decisions for his business.

With advanced forecasting we first try the newest state-of-the-art forecasting algorithms, to improve the currently used forecast. In this first step, we only use the 2 variables mentioned earlier. The second step is adding more variables, such as holiday schedules (3) or the weather forecast (4). The summer holidays may differ from last year, so the number of tourists in that period will also differ. We incorporate the weather forecast as a variable, since the weather will surely influence ice cream sales. With the state-of-the-art forecasting algorithms and the external variables we can improve forecasting accuracy, which gives our ice cream salesman the ability to look further and more accurately into the future, so he knows what to expect. He knows how many raw materials to buy, how to optimize his opening hours, how many workers to hire, etc.