Incorporating Background Knowledge for Better Prediction of Cycle Phases

摘要

When predicting the state of a system, we sometimes know that the succession of states is cyclic. This is for example true for the prediction of business cycle phases, where an upswing is always followed by upper turning points, and the subsequent downswing passes via lower turning points over to the next upswing and so on. We present several ideas of how to implement this background knowledge in popular static classification methods. Additionally, we present a full dynamic model. The usefulness for the prediction of business cycles is investigated, revealing pitfalls and potential benefits of ideas.