Model Selection

In model selection, the best-fitting forecasting model for the given time series is automatically identified, and the corresponding model parameters are optimally adjusted.

Examples of questions that should potentially be answered during model selection include:

• Trend: Is a trend recognizable? If so: How strong is the trend? In which direction does the trend point? Is any trend dampening visible? ersichtlich?
• Seasonality: Are cyclical, recurring patterns visible in the past? If so: How frequently do they occur?
• Adaptivity: How quickly should the forecasting model react to current changes (highly adaptive model), and how much and for how long should past data be incorporated (stable model)?

The competing models can capture and represent different characteristics and structures of the time series in various ways. To compare them, they are fitted to the given time series and examined from different perspectives. An important component in assessing the quality of each model is backtesting.