Components of a Forecasting Model

Question

What are the different components in a forecasting model?

Answer

There are many different ways to produce time series forecasts, and they all involve different mathematical procedures. However, to create a robust forecast it is sensible to look at the existing data and to identify patterns. There are a few standard components that we look for in the data, namely these are; Trend, Seasonal, Cycle and Irregular/Error. This approach is called Classical Decomposition. Trend refers to the overall direction of change in the data, it could be upwards or downwards and can be linear or non-linear. A trendless series, where the mean of the observations stays constant across the period, has ‘stationarity’. Seasonality is change in the values related to the time of year, month or week, with regular peaks and lulls. This is very common in business forecasting with retail often focussed around either Christmas or Summer. Cycle can look similar to Seasonality as it also occurs in wave like peaks and troughs, but occurs over a longer period and may be less regular. An example is the economic cycle. The last component is the irregular component, true irregular components are effectively random, and can’t be predicted. Irregularity causes the jagged appearance of stock market data.

References