Long-term versus short-term forecasting

Long-term strategic planning is different from day-to-day operations planning. In strategic planning, the questions might be “What will be the demand of a particular product in the next year? In the next 5 years? 10 years?” In operations planning, users are more interested in predicting daily/weekly demand. So, the question that often arises is “If I am interested in obtaining forecast data for a specific amount of time in the future, which time bucket should I use?” For example, if you are interested in obtaining forecasts for each of the next three months, you have the following options:

• Use a monthly time bucket in your time series, then predict for the next three time periods.
• Use a weekly time bucket in your time series, then predict for next 12 time periods (weeks) and aggregate them to three monthly forecasts.

In this case, the best option may depend on the number of data points. In a monthly time series with only 10 or 15 data points, weekly forecasting might be a good idea (15 monthly data points means 15 x 4 = 60 weekly data points). However, in a monthly time series of 60 data points (five years of data), monthly forecasting is a better idea. With monthly forecasting, only three future periods must be forecast, while weekly forecasting requires 12 future periods. Each additional point in the future makes forecasting more difficult, affecting the forecast accuracy.