Chapter 7 - Time varying parameter modelling

Abstract

One of the recurrent features of causal tourism demand forecasting models compared with simple time series models, such as the no-change model, has been predictive failure (Witt and Witt, 1995). Predictive failure is normally associated with model structure instability, i.e., the parameters of the demand model vary over time. There are two basic reasons why a causal model may suffer from structural instability. The first, and traditional, view is that our knowledge about the structure of the model is limited. Thus, when a failure occurs we add this information into our knowledge base and produce a better model encompassing both the new and earlier experiences. In this case structural change is regarded merely as a manifestation of an inadequate or inappropriate specification. The second view is that predictive failure and the apparent coefficient changes exhibited in a tourism demand model may well be a reflection of underlying structural change in the data-generating process. This structural change is mainly related to important social, political and economic policy changes. Associated with these two views on the reasons for model instability there are two radically different approaches a modeller could adopt.