財務時間序列模型之估計與檢定方法－－以動差法為基礎的綜合回顧

A Method of Moments Based Synthesis of Estimation and Testing Methods for Financial Time Series Models

本文針對以有限維度動差限制式建構之財務時間序列模型估計與檢定方法，提供一般化的綜合回顧。這些模型包含單變量部分設定架構下的一般化自我迴歸條件異質性模型，及其在單變量完整設定、多變量部分設定（固定／動態條件相關係數模型）、多變量完整設定（以關聯函數為基礎之多變量動態模型）等架構下的延伸模型，與乘式誤差模型。這個綜合回顧的一般性，係來自於一般化殘差的觀念（用以涵蓋不同模型的誤差項）以及動差法（用以得出一般化的估計與檢定）。利用這些基本觀念所整合出的方法，可應用於分析不同的條件動差與條件分配模型。本文藉此整合財務時間序列文獻中一些重要的模型以及對應的參數化估計與檢定方法，並強調這些不同的設定、估計與檢定方法背後具有簡單的通則，可供實證應用者參考。

In this survey paper, we provide a synthesis of parametric (finite-dimensional-momentsbased) estimation and testing methods for various classes of financial time series (FTS) models, including partially specified GARCH-type models in the univariate context and their extensions to univariate fully specified models, multivariate partially specified models (constant /dynamic conditional correlation models), multivariate fully specified models (copula-based multivariate dynamic models), and multiplicative error models. This synthesis is based on a unified approach, which is established using the concept of the generalized residual (that encompasses the error terms of various models) and the method of moments (that forms the generalized estimation and testing methods). This approach is systematically applicable to various conditional moment or distribution models. This paper summarizes a number of important FTS models and the associated parametric estimation and testing methods, and highlights some simple but general principles underlying these seemingly different models and methods.