動態Tobit模式之估計

Abstract

塞車是日常生活中常出現的問題，預測塞車時的車流量成為交通管理上的核心目標；我們希望預測塞車時的車流量，尤其是累積車流量的預測，將有助於交通號誌的設計與控制，進而紓解塞車情況。本文考慮一組實際偵測的交通資料，其中包含車流量與佔有率，針對有截斷（censored）且有延遲（lagged）的特性資料，提出兩種估計方法，第一是針對Poisson迴歸動態Tobit模式採取牛頓法進行參數迭代，第二是針對有潛在獨立變數的動態Tobit模式使用SML-GHK模擬法，模擬潛在變數及概似函數求得參數的最大模擬概似估計式。

Traffic congestion is a serious problem in everybody's daily life. Prediction of traffic flow is becoming the kernel of the traffic management for oversaturated conditions. Our goal is to forecast the traffic flow, especially predict the cumulative traffic flow. This thesis considers the estimation of the parameters of censored data models with lagged dependent variable. Two methods have been proposed. NR algorithm is supplied for solving the Poisson regression dynamic Tobit model. The SML-GHK simulator is another feasible method for the estimation of dynamic Tobit model whose lagged dependent variable is latent.