- 學位論文
間斷性/連續性選擇模型之參數與半參數估計
Semiparametric and Parametric Estimation of Discrete/Continuous Choice Models
摘要
在傳統的間斷性/連續性選擇模型中，隨機效用函數型式及誤差項分配的設定扮演了聯繫間斷性選擇與連續性選擇的重要角色，但這些假設卻可能導致模型設定錯誤的問題。因此，我們嘗試採用兩階段半參數估計法，一方面解除誤差項分配的假設，另一方面仍維持隨機效用函數的設定；在兩階段半參數估計法中，第一階段選擇Klein-Spady （1993）之二元選擇模型半參數估計式，第二階段則採用Newey （1988）的系列函數來近似未知的選擇修正函數。 在實證分析方面，我們以兩階段半參數估計法分析台灣地區家戶之小汽車持有與使用資料，並與傳統之Heckman兩階段估計法進行比較。參數與半參數模型結果皆顯示家戶所得對小汽車持有選擇的影響並不顯著，隱含小汽車已成為一種民生必需品；其次，所有模型皆呈現選擇修正項的顯著性，表示小汽車持有與使用選擇確有相關性存在，與理論模型相符；至於小汽車與機車間係為替代或互補關係，在參數與半參數模型中並無一致的結果。 雖然Klein-Spady模型的結果受到核心函數（kernel function）及其平滑參數（smoothing parameter）選擇的影響，惟以標準常態分配函數作為核心函數、平滑參數0.38之Klein-Spady模型，其估計結果與Probit模型較為接近；一個非正式的圖形檢定結果亦顯示，Probit模型對於資料的描述能力，似乎較Klein-Spady模型為佳。綜合前述分析結果，針對我們的實證資料而言，間斷性/連續性選擇模型之誤差項分配的假設可能相對較不重要。
並列摘要
The assumptions of the functional form of random utility and unobservable error distribution play vital roles in formulating the classical parametric discrete/continuous choice models, serving as the mechanisms for linking the discrete and continuous choices. But the problem of model misspecification that may lead to misleading results arises when incorrect assumptions are made. This study tries to relax the error distributional assumption while retaining the specification of indirect utility function. The two-step semiparametric approach here consists of the Klein-Spady estimator of binary choice model in the first step and Newey’s (1988) series approximation to unknown selectivity correction function in the second step. We compare the traditional Heckman’s two-stage procedure and semiparametric models in application to reanalyzing Liao’s (1996) data on car ownership and use in Taiwan. The comparison of parametric and semiparametric results reveals some important features of the data. First, all of the parametric and semiparametric results are likely to reflect the fact that the car has become a kind of necessity of life for household. Second, the significance of selectivity across models indicates that there exists correlation between car ownership and car use demand, the evidence that is consistent with the theoretical approach. Third, the relationship between the car and the motorcycle becomes ambiguous due to the opposite results obtained from the parametric and semiparametric models. Finally, in the Probit case the predicted probabilities of owning cars with greater index values are much higher than those from Klein-Spady models. Although Klein-Spady results depend on the choices of the kernel function and its bandwidth, the estimated coefficients of Klein-Spady model using a standard normal kernel and setting its bandwidth to 0.38 are relatively close to those from probit model. An informal graphical test result also shows that Probit model seems to describe the data better than does Klein-Spady model. As a result, the specification of the error distribution appears to be less important in discrete/continuous choice model for our data.
參考文獻
延伸閱讀
- 吳國偉（2015）。裝置間通訊下基於機率模型分析之模式選擇演算法〔碩士論文，國立中央大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0031-0412201512095977
- 萬廸翔（2018）。基於常見響應和多項式候選函數之連續時延系統代數鑑別法〔碩士論文，國立成功大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0026-0006202200000112
- 江韋德（2014）。長期追蹤資料下模型選擇方法之比較〔碩士論文，國立中正大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0033-2110201613582471
- Huang, M. Y. (2014). Semiparametric Estimation Approaches for Variant Dimension Reduction Models [doctoral dissertation, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU.2014.02601
- 陳楷文（2014）。Model Selection for Semiparametric Transformation Models under Prevalent Survival Data〔碩士論文，國立清華大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0016-2912201413492245