指導教授：王泓仁臺灣大學：經濟學研究所蘇信瑋Su, Hsin-WeiHsin-WeiSu2014-11-292018-06-282014-11-292018-06-282014http://ntur.lib.ntu.edu.tw//handle/246246/263416The theme of this thesis seeks to find modern testing techniques and estimation methods to support and extend the application of the stochastic frontier models. With a long time development, stochastic frontier (SF, hereafter) models have various homogenous and heterogeneous model specifications, especially on the distribution of the inefficient term. Although many tests in the literature of SF models can help us choose the suitable model specification, these tests can not help us know if we need to use a heterogeneous specification or what kind of heterogeneous specification we should use in SF analysis. Hence, to find a test that can test most kinds of SF models is the first aim of this thesis. On the other hand, with the extension to other econometric fields, the SF analysis requires to use the panel data more frequently. Dynamic panel SF models are models which contains the features of both dynamic panel models and SF models and have the value in SF analysis when using panel data. However, this kind of models are more difficult to estimate than either dynamic panel models or SF models. To seek a way to consistently estimate most kinds of dynamic panel SF models is the second aim of this thesis. In this thesis, three chapters are generated to discuss the aforementioned testing and estimating issues in SF models: 1. Evaluating Stochastic Frontier Models by the Simulated Integrated Conditional Moment Test The problem of testing the distribution of the composite error or the functional form of the frontier function in the SF models has become increasingly important in recent years. However, the tests mentioned in the literature of SF analysis are not able to jointly test the misspecification of different aspects of SF models, especially the distribution of the composite error and the functional form of the frontier function. The lack of appropriate tests may lead to incorrect model specifications for empirical analysis. This paper applies the SICM test of Bierens and Wang (2012) to SF models.The SICM test is a consistent test with p n non-trivial local power, and it can detect comprehensively the misspecification of many aspects of the model. This paper also demonstrates the validity and advantages of this test in practical applications using a i Monte Carlo simulation. 2. Moment Estimators for Dynamic Panel Stochastic Frontier Models with Fixed-Effect SF models have widely applied in more and more econometric fields, but this extension brings new questions and challenges. When analyzing the panel data with some dynamic property, the researchers now may encounter a dynamic panel SF model which means there is the incidental parameter problem caused by an unobserved individual variable and the lagged terms of the dependent variable in the production function of the SF models. This chapter tries to find an estimation strategy which may consistently estimate the parameters of the dynamic panel SF model. Referring to Chen and Wang (2014), the estimation strategy contains two step. The first step applies the dynamic panel generalized method of moments (GMM, hereafter) method to estimate the parameters of the production function. In the second step, we use the method of moments to obtain the moment estimators of the distribution parameters of the composite error. The simulation results demonstrate that these estimators may be consistent when the numbers of individual go to infinity. 3. Quasi-Maximum Likelihood Estimation for Heteroscedastic Dynamic Panel Stochastic Frontier Models SF models with heteroscedastic composite errors can analyze the factors that influence the inefficient term and have become highly popular recently. This chapter succeeds the work of the second to find a consistent estimation method for dynamic SF model with heteroscedastic composite errors. Two-step approach is still be adopted but with some changes for estimating heteroscedastic dynamic panel SF models. In the first step, the dynamic panel GMM technique is still be applied to estimate the parameters of production function, but in the second step, the quasi-maximum likelihood (QML, hereafter) estimation is conducted to obtain the distribution parameter estimators of the composite error. The identification of QML estimation confirms the consistency of this method. The simulation results also illustrate that the estimators of heteroscedastic dynamic panel SF models are consistent when one uses this two-step approach.Table of Contents 1 Introduction 1 2 Evaluating Stochastic Frontier Models by the Simulated Integrated Conditional Moment Test 3 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2.2 The SICM test for the SF models . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.3 Other specification tests in the SF model literatures versus SICM test . . . . . . 10 2.4 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.4.1 The finite sample size of SICM test . . . . . . . . . . . . . . . . . . . 15 2.4.2 The finite sample power of SICM test . . . . . . . . . . . . . . . . . . 19 2.5 An empirical study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 3 Moment Estimators for Dynamic Panel Stochastic Frontier Models with Fixed- Effect 26 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 3.2 Introduction to the dynamic panel GMM method . . . . . . . . . . . . . . . . 28 3.3 The strategy to estimate the dynamic panel SF model . . . . . . . . . . . . . . 30 3.3.1 Using Dynamic panel GMM to estimate the production function . . . . 30 3.3.2 The moments estimator for the dynamic panel SF models . . . . . . . . 33 3.3.3 Asymptotic covariance of moment estimators . . . . . . . . . . . . . . 37 3.4 The simulation of moment estimators for the dynamic panel SF model . . . . . 40 3.5 The conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 4 Quasi-Maximum Likelihood Estimation for Heteroscedastic Dynamic Panel Stochastic Frontier Models 55 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 4.2 Dynamic panel GMM estimation for the production function . . . . . . . . . . 58 4.3 Quasi-maximum likelihood estimation of parameters in composite errors . . . . 59 4.3.1 Identification of QML estimation . . . . . . . . . . . . . . . . . . . . 61 iii 4.3.2 The asymptotic estimators of QML estimators . . . . . . . . . . . . . . 63 4.4 Simulation for heteroscedastic dynamic panel SF models . . . . . . . . . . . . 66 4.5 An empirical study: estimation of growth convergence by using a dynamic panel stochastic frontier model . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773938506 bytesapplication/pdf論文公開時間：2015/08/17論文使用權限：同意有償授權(權利金給回饋學校)隨機邊界模型模型設定檢定動態追蹤模型兩階段估計動差法準最大概似估計thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/263416/1/ntu-103-D96323004-1.pdf