| Summary: | 碩士 === 國立臺北大學 === 統計學系 === 92 === Pattern identification methods have been widely used for determining the order of an ARMA process, such as S-Array and R-Array Method、GPAC Method and ESACF Method. Compared with penalty function identification methods, the pattern identification methods are computationally cheap. Corner method is also one of the pattern identification methods. Most important, it can be used for transfer-noise function model.
Corner method was proposed by Beguin, Gourieroux, and Monfort in 1980. In 1982, Liu and Hassens used the least square method to find the estimates of the regression coefficients and produce an array to determine the order of a multiple-input transfer function. It still has the advantages mentioned above. However, Liu and Hassens neither proved the patterns of the array nor proposed a test procedure on their article. This study wants to know more about the unknown statistical properties of corner method.
Since the asymptotic distribution of the test statistics of the array is difficult to be derived, we use repeating simulation to find the variance of the elements and use them to be variance estimates to test each element in the array. Given different parameters、numbers of simulation and error term models, we inspect that if the results of the simulation and the theory are coincident. Also, the patterns of the array are also proved in the research.
According to the results of the simulation, we find that it almost can determine the true model in different kinds of situations, which means the variance calculated from repeating simulation is well-behaved. Although it seems to be over estimated in the tail of the array, the results of the model identification aren’t affected too much.
|
|---|
