Estimating the effects of changes in pricing strategies on a company’s sales or market share is important task faced by managers. The intervention analysis, a time-series technique initially proposed by Box and Tiao, has been utilized for this purpose. However, previous applications of the intervention analysis have been limited to univariate time-series data.
In this thesis, we propose a new intervention model which can be applied to multiple time-series data (panel data) and a forecasting procedure. We propose Bayesian hierarchical estimations and pooled estimations using the model. By borrowing strength across cross-sectional units, theses estimation schemes give more robust and reasonable result than one from the individual estimation. Furthermore, the proposed schemes yield improved predictive power in the forecasting of hold-out sample periods.
