- 學位論文
Estimating Taiwan's Monthly GDP in an Exact Kalman Filter Framework
HASH(0x1cbeb830)
摘要
This dissertation aims to compare the results of the state-space model without the co-integration assumption which is a special case of Mitchell et al. (2005) along with the different disaggregation approaches using Taiwan's data. The motivation behind disaggregation, the various fields that disaggregation applied to, and the necessity of disaggregation in data collection are introduced briefly. Three streams of temporal disaggregation are discussed in the literature, namely the mathematical approach, the statistical dynamic approach and finally the state-space approach by which this dissertation gives emphasis on. The state-space approach which encompasses the two approaches and considers the underlying dynamic structure turns out to be the most general flexible approach. It is illustrated in this dissertation how a state-space model is set up to disaggregate quarterly real GDP into monthly estimates and carries on the filtering and smoothing procedure of exact Kalman filter of the non-stationary case. To compare the state-space model result with the published quarterly GDP series, an annually to quarterly estimate is performed. Furthermore, the empirical results show the in-sample fitting and out-of-sample forecast for the quarterly to monthly disaggregated real GDP. It can be clearly shown that the in sample fitting results of state-space model and Santos Silva and Cardoso exhibit relatively smooth monthly GDP estimates which match with our intuition. Yet, in the out-of-sample forecast, state-space form without co-integration assumption is the only model with the smallest root mean squared error measure. Further research could be developed on constructing a state-space Markov switching model to restore the characteristics of monthly disaggregated GDP. Thus, the Markov switching property could be preserved when aggregated back into quarterly GDP series providing more details in the disaggregation transformation process.