經濟理論的動態最適化架構了所謂的動差條件模型。一般化動差估計法~(GMM)， 是目前文獻上相當普遍使用的動差條件估計技術。 然而，在許多的模擬與實証研究已發現， GMM 在小樣本的表現上不盡理想: 點估計式存在著嚴重的偏誤~(bias)，以及相關的檢定統計量具有相當的型一誤差扭曲現象。 在本篇論文當中，我們將探討一個由無母數概似法所發展出來的方法，empirical likelihood 架構於動差條件模型之使用。 本文主要有底下兩個特色: 在動差條件模型的架構之下， 一、 我們完整的闡釋了 empirical likelihood 估計與統計推論的理論性質， 並且我們的討論重點主要將著眼於與既有的 GMM 估計式做比較性探討。 二、 透過蒙地卡羅模擬，我們試驗了幾種計量模型，討論 empirical likelihood 點估計式與其相關的檢定在小樣本上的表現，並與文獻上其他各種的動差條件估計式做比較分析。 從我們大部分的模擬結果可發現，傳統上的 GMM 估計式並不能提供令人滿意的小樣本表現; empirical likelihood 估計式可以提供相當準確的小樣本點估計與較可信的統計推論結果。 Moment condition models arise naturally from the dynamic economic theory with optimizing agents. The generalized method of moments (GMM) estimation proposed by Hansen (1982) has been a popular estimation technique for moment condition models in the literature. However, many Monte Carlo and empirical evidences found that the GMM estimator may be severely biased and the associated tests may have substantial size distortions in small samples. In this thesis, we explore a method originally developed in nonparametric likelihood framework. The usefulness of the empirical likelihood estimation and inferences are investigated under unconditional moment condition models. In particular, we focus on the over-identified moment condition models. Two emphases are comprehended in the thesis. First, we clarify the theoretical aspects of empirical likelihood, including both estimation and tests. Our emphasis is specifically put on the comparisons with the conventional GMM framework. Second, using Monte Carlo simulations we examine the small-sample performances of the empirical likelihood estimator and compare with several competitive estimators in different well-known econometric models. In most of our Monte Carlo experiments, we confirm the poor small-sample performances of the conventional GMM estimator, and the empirical likelihood estimator can provide less biased estimates and more reliable inferences in small samples.
