您的瀏覽器不支援Javascript,部分功能將無法呈現。

:::

例行學術演講
  • 標題:2018 年 12 月 21 日CRETA Seminar @TES
  • 公告日期:2018-11-28

 

CRETA Seminar

日期:2018  12  21  (週五下午2:00~5:00

地點:國立臺灣大學管理學院二號館三樓 304 教室

講者:楊睿中教授(國立清華大學經濟學系)

演講主題:Tree, Random Forest, Gradient Boosting, Double Machine Learning, and a Discussion on Propensity Score Matching

 

講題摘要:

  The gradient boosting and the random forest have been widely used in empirical applications. Computationally, the gradient boosting performs the functional gradient descent on the loss function by repeatedly fitting a weak learner, typically a shallow classification and regression tree (CART), to the residuals. However, since CART is a special case of multivariate adaptive regression splines (MARS), gradient boosting with shrinkage can be viewed as an infinitesimal forward-stagewise spline regression, which is a special version of a (constant) spline regression with a L1 penalty, a.k.a., the least absolute shrinkage and selection operator (LASSO). On the other hand, the random forest is a combination of the bootstrap aggregating (bagging) and the CART. Nevertheless, it has been shown that the random forest can also be viewed as an adaptively weighted k potential nearest neighbors (k-PNN) method.

Recently Chernozhukov et al. (2018) proposed a new double/debiased machine learning frame-work  (DML)  for  the  estimation  and  inference  of  low-dimensional  parameters  in  the  presence  of high-dimensional nuisance parameters. In DML, the low-dimensional parameters of interest are estimated using the Neyman orthogonal moment and the cross-fitting technique, while the nuisance parameters are estimated by some machine learning algorithms with sufficient rates of convergence.  We show that a) regression trees and random forests in general do not have the sufficient rates and may result in serious bias and size distortion, and b) when unknown nuisance functions are additive, the gradient boosting with stumps provide consistent estimation and correct inference for the treatment effect.

We apply our methods to recent debate of the treatment effect of the Big N auditors to the audit quality. Consistent to the result of DeFond et al. (2016), who use 3,000 designs of the propensity score matching, our method supports the existing of the Big N effect during 1988 to 2006 in U.S. DML-GB also identifies the non-linear associations of the firm characteristics and the audit quality. 

講者介紹:

楊睿中教授 (國立清華大學經濟學系助理教授,個人網頁:https://sites.google.com/site/juichungyang/)

 

會程安排:

下午 130 ~ 200報到

下午 200 ~ 320 First session

下午 320 ~ 340 Tea Break 

下午 340 ~ 500 Second session

為方便場地安排及人數預估,欲參加CRETA Seminar的朋友們,煩請事先報名。

當天講義將優先提供給報名者

報名網址:http://www.creta.org.tw/?news_2=214    

報名費用:台灣大學在學學生及現任教職員和台灣經濟計量學會會員為免費參加

其他參加者報名費為NT$200 (當天將開放現場繳交台灣經濟計量學會 2019 年年度會費)

報名期限:2018/12/20 () 13:30

歡迎各位踴躍參加!!

為方便臺灣經濟計量學會 (TES) 會員繳納 108 年度會費,本次活動開放現場繳納會費,亦歡迎大家介紹非會員朋友加入 TES。更多研討會資訊請見 TES 網站:http://www.tesociety.org.tw/main.php

此外,臺灣經濟計量學會新開設臉書粉絲專頁,歡迎各位舊雨新知按讚及追蹤我們各項活動的最新訊息!網址如下:

https://www.facebook.com/twtes/

  • 附件檔案:
  • 參考連結:
  • 張貼人:網站管理員
  • 最後修改時間:2018-11-28 PM 7:34

web_use_log