以計算實用之模擬估計法所估計之動態面板數據Tobit模型

A Computationally Practical Simulation Estimation for Dynamic Panel Tobit Models

在考慮動態面板數據Tobit模型多種可能的相關結構下，本文提出一個容易計算並實用的模擬估計法。此模擬估計法式藉由相關隨機估計法來執行。其取對數之概似函數是經由GHK及Gibbs模擬方法加以模擬並求其極大值。動態模型中之起始值問題在本文中亦有詳細之討論。Scholz et al. (2006)曾使用本文所提出的模擬估計法是估計動態面板數據Tobit模型並應用於美國退休金問題之探討。本文所作之模擬實驗亦顯示所提出之動態面板數據Tobit模型之模擬估計法在模擬樣本數小的情況下亦非常精確。

This paper provides a computationally practical simulation estimation for the dynamic panel Tobit model with large categories of dependence structures. The simulation estimators are conducted through correlated random effects approach. The log-likelihood function is simulated and maximized through procedures based on a recursive algorithm formulated by GHK and Gibbs sampling simulators. The initial conditions problem is discussed in details. The simulation estimators discussed in this paper have been implemented by Scholz et al. (2006) to study retirement savings. Monte Carlo experiments indicate that the simulation estimators perform strikingly well, especially in the case of the Gibbs estimator, even for a small simulation size.