台灣汽車車體險理賠因素分析:受限制分量迴歸之應用

An Analysis of Incurred Losses of Physical Damage Insurance in Taiwan: An Application of Censored Quantile Regression

莊碩立(Shuo-Li Chuang)

指導教授 ： 管中閔

國立臺灣大學/管理學院/財務金融學研究所/碩士(2004年)

https://doi.org/10.6342/NTU.2004.01204

摘要

受限制分量迴歸可對受限制資料提供分量上的分析,
但是並不容易估計係數。
三步法是用來計算受限制分量迴歸估計值的一種簡易方法,
它特別適用在當大部分資料受到限制,
並且包含釵h類別變數的情況,
而保險公司的理賠金額資料就具有這兩種特性。
台灣的保險公司最常使用的最小平方法,
因為忽略理賠金額資料之受限制性質,
以及僅提供平均數的分析,
可能無法對理賠金額提供準確的分析。
本文使用三步法,
其估計結果與最小平方法及Tobit模型相當不同,
故對不同分量下的理賠金額提供了不同的解釋。

關鍵字

受限制分量迴歸 ； 三步法 ； 車體險損失 ； 分量迴歸

並列摘要

The censored quantile regression (CQR) provides a quantile analysis on
censored data, but its objective function is
difficult to solve. The 3-step method for the CQR estimation is a simple
method to compute the regression estimates. This method is particularly
suitable for quantile analysis when data are highly cnesored and contains many
categorical regressors.
The amount of incurred loss of an insurance company is a kind of data with these two properties.
The insurers in Taiwan may not always make the precise prediction of the amount
of incurred loss since they apply only the ordinary least squares (OLS) method, which ignores censoring values and
olny analyzes the mean behavior of data.
In this thesis, the 3-step method for CQR estimation provides different explanations on
the amount of incurred loss at distinct quantiles.
This results are quite different from those based on
OLS and Tobit estimation.

並列關鍵字

quantile regression ； censored quantile regression

參考文獻

and Related Method, ed. Y. Dodge, Hayward, California. IMS.

Bilias, Y., S. Chen and Z. Ying (2000). Simple resampling methods for censored regression

Buchinsky, M. (1991). Changes in the U.S. wage structure 1963-1987: application of quantile regression,

Econometrica 62, 405-458.

Buchinsky, M. (1998). Recent advances in quantile regression models: a pratical guideline for