Statistical Methods for the Seasonal Adjustment

本研究的主题如下：1.信息标准EIC的使用：讨论了扩展信息标准EIC在季节调整中的应用。结果表明，本文提出的重采样方案具有选择较好趋势进行样本外预测的自然趋势。多变量经济序列分析：本研究旨在分析会话调整程序作为多变量时间序列分析的一种前处理技术的作用。研究发现，当对每个序列分别进行季节性和去趋势性调整时，存在丢失相关序列之间相互关系的信息的危险。X-11型模型：我们尝试用基于现代模型的季节调整方法重建X-11型趋势估计。我们介绍了一种与传统的基于加性模型的趋势估计相结合的偏差修正方法。我们还介绍了一种具有X-11型趋势的新机型，用于多路…更多的VE系列。使用新模型，我们可以直接从数据中估计X-11型趋势。蒙特卡罗滤波：发展了一种蒙特卡罗滤波和平滑方法，用于高维非线性非高斯状态空间模型的状态估计。基于这种方法，考虑了季节调整的各种模型，例如：(1)趋势或季节分量跳跃的检测(2)异常值的处理(3)乘性模型的估计(4)超参数的贝叶斯估计。遗传算法：我们研究了遗传算法和蒙特卡罗滤波之间的关系。本文的主要目的是从蒙特卡罗滤波的角度对遗传算法进行解释，将其引入贝叶斯框架。分解的改进：提出了一种提取平稳平稳自回归分量的方法。其中我们考虑了一个在频域内有约束的数值优化问题。然而，进一步的研究仍未完成。特别是在造型上跳跃和扭结在潮流中。协整模型：提出了协整模型的状态空间表示法，使我们能够一步估计未知参数。而传统的Engle-Granger方法需要两个步骤。X-12-REGARIMA：提高其预测能力。基于AIC的回归模型选择程序被纳入传统的基于移动平均的X-11季节调整程序中。较少

Topics covered in this study are as follows :1. Use of Information Criterion EIC : Use of Extended Information Criterion EIC for the seasonal adjustment is discussed. It is demonstrated that the proposed resampling scheme shows a a natural tendency for chosing better trend for the out-of-sample forcasting.2. Multivariate Economic Series Analysis : This study intended to analyze the role of sesonal adjustment procedure as a preprocessing technique for the multivariate timeseries analysis. It is reveald that there is danger of losing information about the mutual relationship among related series, when each series is adjusted for the seasonality and detrended separately.3. X-11 type model : We tried to reconstruct the X-11 type trend estimate by the modern model-based seasonal adjustment method. We introduced a bias correction method which is to be used with conventional additive type model-based trend estimate. We also introduced a new model which has the X-11 type trend for multiplicati … More ve series. With the new model we can estimate the X-11 type trend directly from the data.4. Monte Carlo filtering : A Monte Carlo filtering and smoothing methods have been developed for state estimation of high-dimensional nonlinear non-Gaussian state space models. Based on this methods, various models for seasonal adjustment are considered, e. g. , (1) detection of jumps of trend or seasonal components (2) treatment of outliers (3) estimation of multiplicative model (4) Baysian estimation of hyper-parametrs.5. Genetic Algoritm : We investigate the relationships between the Genetic Algoritm and Monte Calro Filter. The major objective of this paper is to cast the Genetic Algorithm into the Baysian framework by its interpretation from a viewpoint of the Monte Carlo filter.6. Improve of DECOMP : A procedure for extracting 'stable' stationary autoregressive component is proposed. in which we consider a numerical optimization with a restriction in frequency domain. Further research remained undone, however. especially in modeling jump and kink in trend.7. Co-integration model : State-space representation for co-integration model is proposed which enables us to estimate the unknown parameters in one-step. while traditional Engle-Granger's method needs two steps.8. X-12-REGARIMA : To improve its forcasting ability. AIC based regression model selection procedure is incorporated in the traditional moving-avarage based X-11 seasonal adjustment program. Less