사업팀 구성 nonparametric and semiparametric function estimation 비모수적, 준모수적...

1
사사사 사사 Nonparametric and Semiparametric Function Esti mation 사사사사 , 사사사사 사사사사 사사사 ( 사사사사사 사사사사 ) http://stats.snu.ac.kr/~brain03 Nonparametric Stochastic Frontier - A first approach to fit nonparametric stochastic models by local maximum likelihood techniques - The models encompass anchorage parametric models in a no nparametric way and have convoluted error terms (efficienc y plus noise) - Derive asymptotic properties of the estimators of the st ochastic frontier models - Establish the performance of the estimators with simulat ed data sets and with a real data on US commercial banks - The methods are robust, numerically stable and particula rly useful for investigating production processes and the derived efficiency scores - To be published in Journal of Econometrics (2006) Bivariate example with heteroscedasticity: True against estimated f rontier (left panel) and true against predicted efficiencies (right panel). The asterisks (blue) correspond to the local MLE and the circles (red) to the MLE of the parametric anchorage model. Asymptotic Theory for ARCH-SM Models : LAN and Residual Processes - Propose efficient estimator and model checking procedure in GARCH(p,q) model by investigating LAN property in ARCH (∞) model and the limiting distribution of residual empir ical process - Published in Statistica Sinica (2005) 사사사사 박박 : 박박박 (2004), 박박박 (2005), 박박박 (2005), 박박박 (2005) 박박 : 박박박 (2004), 박박박 (2004), 박박박 (2004), 박박박 (2005) 사사사사 Professor Byeong Uk Park was elected 2005 Fellow of Americ an Statistical Association and also 2005 Fellow of Institu te of Mathematical Statistics, for his outstanding contrib utions to the statistical profession. 사사사사 : 박박박 사사사사 : 박박박 , 박박박 , 박박박 , 박박박 , 박박박 사사사사사사 : 박박박 , 박박박 , 박박박 , 박박박 , 박박박 , 박박박 , 박박박 , 박박박 박박박 ( 박박 박박박박 ), 박박박 , 박박박 ( 박박 박박박박 ) 사사사사사사 : 박박박 , 박박박 SCI 사사사 (2003 사 사사 ) : 52 사 사사사사사사 : Bandwidth Selection for Smooth Backfitting in Additive Models - Propose three fully automated bandwidth selection methods for the smooth backfitting in additive models which is know n to achieve the oracle efficiency bound - Derive uniform higher-order stochastic expansions for the residual sums of squares of the smooth backfitting estimato rs - Give the large sample properties of the bandwidth selecti on methods and compare their finite sample properties throu gh simulation experiments - Published in The Annals of Statistics (2005) A Simple Smooth Backfitting in Additive Models - Propose a new smooth backfitting estimator for additive n onparametric regression models - The estimator has the simple structure of Nadaraya-Watson smooth backfitting and achieves the oracle property of loca l linear smooth backfitting - Each component function is estimated with the same asympt otic accuracy as if the other components would be known - To be published in The Annals of Statistics (2006) LAN property ARCH(∞) model : Empirical process Efficient estimator Goodness of fit Model check , t t t y ) 1 , 0 ( ~ iid t 1 2 1 0 2 j j t j t y 사사사 사사 Higher order expansion of n i d j i j j i i X m m Y X w n h RSS 1 1 2 0 )} ( ˆ ˆ ){ ( 1 ) ( Higher order expansion of n i d j d j i j j i j j i X m m X m m X w n h ASE 1 1 2 1 0 0 } ) ( ) ( ˆ ˆ ){ ( 1 ) ( d j j K nh h RSS h PLS 1 )} 0 ( 1 2 1 ){ ( ) ( ) ( min arg h ASE h opt Bandwidth selection by minimizing PLS(h) Bandwidth selection by estimating h opt Bandwidth selection by estimating h opt,j ) ( min arg , j j j opt h ASE h Smooth backfitting additive model : d j i i j j i X m m Y 1 0 ) ( ) ( ˆ , ), ( ˆ , ˆ 1 0 d m m m

Upload: ashlynn-brooks

Post on 13-Dec-2015

224 views

Category:

Documents


6 download

TRANSCRIPT

Page 1: 사업팀 구성 Nonparametric and Semiparametric Function Estimation 비모수적, 준모수적 함수추정 연구팀 ( 서울대학교 통계학과 ) brain03 Nonparametric Stochastic

사업팀 구성

Nonparametric and Semiparametric Function Estimation비모수적 , 준모수적 함수추정 연구팀 ( 서울대학교 통계학과 )

http://stats.snu.ac.kr/~brain03

Nonparametric Stochastic Frontier

- A first approach to fit nonparametric stochastic models by local maximum likelihood techniques - The models encompass anchorage parametric models in a nonparametric way and have convoluted error terms (efficiency plus noise)- Derive asymptotic properties of the estimators of the stochastic frontier models- Establish the performance of the estimators with simulated data sets and with a real data on US commercial banks- The methods are robust, numerically stable and particularly useful for investigating production processes and the derived efficiency scores - To be published in Journal of Econometrics (2006)

Bivariate example with heteroscedasticity: True against estimated frontier (left panel) and true against predicted efficiencies (right panel). The asterisks (blue) correspond to the local MLE and the circles (red) to the MLE of the parametric anchorage model.

Asymptotic Theory for ARCH-SM Models : LAN and Residual Processes

- Propose efficient estimator and model checking procedure in GARCH(p,q) model by investigating LAN property in ARCH(∞) model and the limiting distribution of residual empirical process- Published in Statistica Sinica (2005)

인력배출박사 : 이영경 (2004), 유규상 (2005), 남승민 (2005), 윤영주 (2005)석사 : 송효임 (2004), 이미희 (2004), 최혜정 (2004), 정옥경 (2005)

기타성과Professor Byeong Uk Park was elected 2005 Fellow of American Statistical Association and also 2005 Fellow of Institute of Mathematical Statistics, for his outstanding contributions to the statistical profession.

Picture taken after IMS Awards Session, the 165 Annual Joint Meeting, Minneapolis, Minnesota, U.S.A., August 7-11, 2005

사업팀장 : 박병욱

참여교수 : 김우철 , 송문섭 , 이상열 , 전종우 , 조신섭

참여대학원생 :

권성훈 , 김주원 , 송대건 , 연규필 , 여동화 , 이지연 , 이태욱 , 하소영황준혁 ( 이상 박사과정 ), 김지영 , 정보혜 ( 이상 석사과정 )

신진연구인력 : 이영경 , 김영진

SCI 논문수 (2003 년 이후 ) : 52 편

대표연구업적 :

Bandwidth Selection for Smooth Backfitting in Additive Models

- Propose three fully automated bandwidth selection methods for the smooth backfitting in additive models which is known to achieve the oracle efficiency bound- Derive uniform higher-order stochastic expansions for the residual sums of squares of the smooth backfitting estimators- Give the large sample properties of the bandwidth selection methods and compare their finite sample properties through simulation experiments- Published in The Annals of Statistics (2005)

A Simple Smooth Backfitting in Additive Models

- Propose a new smooth backfitting estimator for additive nonparametric regression models - The estimator has the simple structure of Nadaraya-Watson smooth backfitting and achieves the oracle property of local linear smooth backfitting- Each component function is estimated with the same asymptotic accuracy as if the other components would be known - To be published in The Annals of Statistics (2006)

True function and observed data (left panel); Smooth backfiiting estimate (right panel)

LAN property

ARCH(∞) model :

Empirical process

Efficient estimator Goodness of fit Model check

,ttty )1,0(~ iidt

1

2

10

2

j jt

j

t y

사업팀 성과

Higher order expansion of

n

i

d

j

ijj

ii XmmYXwn

hRSS1 1

20 )}(ˆˆ){(

1)(

Higher order expansion of

n

i

d

j

d

j

ijj

ijj

i XmmXmmXwn

hASE1 1

2

100 })()(ˆˆ){(

1)(

d

j j

Knh

hRSShPLS1

)}0(1

21){()( )(minarg hASEhopt

Bandwidth selection by minimizing PLS(h)

Bandwidth selection by estimating hopt

Bandwidth selection by estimating hopt,j

)(minarg, jjjopt hASEh

Smooth backfitting additive model :

d

j

iijj

i XmmY1

0 )(

)(ˆ,),(ˆ,ˆ 10 dmmm