Binomial Distribution: Tossing a coin m times. p = probability of having head from a trial. y = # of having heads from n trials (y = 0, 1,..., m).

Similar documents
Maximum Likelihood Estimation of Binary Dependent Variables Models: Probit and Logit. 1. General Formulation of Binary Dependent Variables Models

Maximum Likelihood Estimation of Binary Dependent Variables Models: Probit and Logit. 1. General Formulation of Binary Dependent Variables Models

Probability and Random Variable Primer

CS 2750 Machine Learning. Lecture 5. Density estimation. CS 2750 Machine Learning. Announcements

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X

Statistics Spring MIT Department of Nuclear Engineering

Econ107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)

Chapter 20 Duration Analysis

Machine learning: Density estimation

j) = 1 (note sigma notation) ii. Continuous random variable (e.g. Normal distribution) 1. density function: f ( x) 0 and f ( x) dx = 1

Marginal Effects in Probit Models: Interpretation and Testing. 1. Interpreting Probit Coefficients

See Book Chapter 11 2 nd Edition (Chapter 10 1 st Edition)

Generative classification models

Suites of Tests. DIEHARD TESTS (Marsaglia, 1985) See

First Year Examination Department of Statistics, University of Florida

Exam. Econometrics - Exam 1

Limited Dependent Variables

Hydrological statistics. Hydrological statistics and extremes

Introduction to the R Statistical Computing Environment R Programming

Chapter 2 - The Simple Linear Regression Model S =0. e i is a random error. S β2 β. This is a minimization problem. Solution is a calculus exercise.

Using T.O.M to Estimate Parameter of distributions that have not Single Exponential Family

e i is a random error

Copyright 2017 by Taylor Enterprises, Inc., All Rights Reserved. Adjusted Control Limits for P Charts. Dr. Wayne A. Taylor

MATH 281A: Homework #6

The Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction

MATH 829: Introduction to Data Mining and Analysis The EM algorithm (part 2)

Maximum Likelihood Estimation

Diagnostics in Poisson Regression. Models - Residual Analysis

Properties of Least Squares

β0 + β1xi and want to estimate the unknown

Simulation and Random Number Generation

Random Partitions of Samples

A New Method for Estimating Overdispersion. David Fletcher and Peter Green Department of Mathematics and Statistics

Predictive Analytics : QM901.1x Prof U Dinesh Kumar, IIMB. All Rights Reserved, Indian Institute of Management Bangalore

SELECTED PROOFS. DeMorgan s formulas: The first one is clear from Venn diagram, or the following truth table:

The EM Algorithm (Dempster, Laird, Rubin 1977) The missing data or incomplete data setting: ODL(φ;Y ) = [Y;φ] = [Y X,φ][X φ] = X

CIS526: Machine Learning Lecture 3 (Sept 16, 2003) Linear Regression. Preparation help: Xiaoying Huang. x 1 θ 1 output... θ M x M

Statistical analysis using matlab. HY 439 Presented by: George Fortetsanakis

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)

Primer on High-Order Moment Estimators

Chapter 1. Probability

Engineering Risk Benefit Analysis

Lecture 3 Specification

Parametric fractional imputation for missing data analysis. Jae Kwang Kim Survey Working Group Seminar March 29, 2010

Applied Stochastic Processes

Lecture 3: Probability Distributions

However, since P is a symmetric idempotent matrix, of P are either 0 or 1 [Eigen-values

MIMA Group. Chapter 2 Bayesian Decision Theory. School of Computer Science and Technology, Shandong University. Xin-Shun SDU

Probability Theory (revisited)

Topic 5: Non-Linear Regression

BIO Lab 2: TWO-LEVEL NORMAL MODELS with school children popularity data

+, where 0 x N - n. k k

b ), which stands for uniform distribution on the interval a x< b. = 0 elsewhere

since [1-( 0+ 1x1i+ 2x2 i)] [ 0+ 1x1i+ assumed to be a reasonable approximation

Lecture Notes on Linear Regression

Copyright 2017 by Taylor Enterprises, Inc., All Rights Reserved. Adjusted Control Limits for U Charts. Dr. Wayne A. Taylor

The corresponding link function is the complementary log-log link The logistic model is comparable with the probit model if

xp(x µ) = 0 p(x = 0 µ) + 1 p(x = 1 µ) = µ

Chapter 4: Regression With One Regressor

Tests of Single Linear Coefficient Restrictions: t-tests and F-tests. 1. Basic Rules. 2. Testing Single Linear Coefficient Restrictions

Chapter 2 Transformations and Expectations. , and define f

APPROXIMATE PRICES OF BASKET AND ASIAN OPTIONS DUPONT OLIVIER. Premia 14

Course 395: Machine Learning - Lectures

ANSWERS CHAPTER 9. TIO 9.2: If the values are the same, the difference is 0, therefore the null hypothesis cannot be rejected.

Systems of Equations (SUR, GMM, and 3SLS)

The conjugate prior to a Bernoulli is. A) Bernoulli B) Gaussian C) Beta D) none of the above

1. Inference on Regression Parameters a. Finding Mean, s.d and covariance amongst estimates. 2. Confidence Intervals and Working Hotelling Bands

XII.3 The EM (Expectation-Maximization) Algorithm

Decision-making and rationality

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Limited Dependent Variables and Panel Data. Tibor Hanappi

Logistic regression models 1/12

Econ Statistical Properties of the OLS estimator. Sanjaya DeSilva

Chapter 7 Generalized and Weighted Least Squares Estimation. In this method, the deviation between the observed and expected values of

JAB Chain. Long-tail claims development. ASTIN - September 2005 B.Verdier A. Klinger

Lecture 4 Hypothesis Testing

Chapter 3. Two-Variable Regression Model: The Problem of Estimation

Lab 4: Two-level Random Intercept Model

Distributions /06. G.Serazzi 05/06 Dimensionamento degli Impianti Informatici distrib - 1

Classification learning II

( ) [ ( k) ( k) ( x) ( ) ( ) ( ) [ ] ξ [ ] [ ] [ ] ( )( ) i ( ) ( )( ) 2! ( ) = ( ) 3 Interpolation. Polynomial Approximation.

Linear Approximation with Regularization and Moving Least Squares

Implementation of the Matrix Method

STAT 405 BIOSTATISTICS (Fall 2016) Handout 15 Introduction to Logistic Regression

Propagation of error for multivariable function

ECONOMICS 351*-A Mid-Term Exam -- Fall Term 2000 Page 1 of 13 pages. QUEEN'S UNIVERSITY AT KINGSTON Department of Economics

Fall 2012 Analysis of Experimental Measurements B. Eisenstein/rev. S. Errede

MACHINE APPLIED MACHINE LEARNING LEARNING. Gaussian Mixture Regression

A be a probability space. A random vector

10-701/ Machine Learning, Fall 2005 Homework 3

Expectation Maximization Mixture Models HMMs

Hidden Markov Models & The Multivariate Gaussian (10/26/04)

Expected Value and Variance

INF 5860 Machine learning for image classification. Lecture 3 : Image classification and regression part II Anne Solberg January 31, 2018

Math 426: Probability MWF 1pm, Gasson 310 Homework 4 Selected Solutions

MLE and Bayesian Estimation. Jie Tang Department of Computer Science & Technology Tsinghua University 2012

Randomness and Computation

The Ordinary Least Squares (OLS) Estimator

Modeling and Simulation NETW 707

Exercises of Chapter 2

Transcription:

[7] Count Data Models () Some Dscrete Probablty Densty Functons Bnomal Dstrbuton: ossng a con m tmes p probablty of havng head from a tral y # of havng heads from n trals (y 0,,, m) m m! fb( y n) p ( p) p ( p) y y!( m y)! E(y) mp; var(y) mp(-p) y y y m y Posson Dstrbuton: Let p µ/m for a bnomal dstrbuton y µ µ f p( y) lm m fb( y m) e, y 0,,, y! E(y) var(y) µ LDV-38

Dgresson to the Chocolate Chp Cookes problem: Wsh to put at least one CC on a cooke wth 95% he CC machne locates CC s n a cooke followng a Posson Dstrbuton wth µ Pr(y 0) < 005 e -µ < 005 µ > -ln(005) 995 You need a machne wth µ at least 3 Geometrc Dstrbuton: y # of trals untl havng a head f ( y) ( p) y p, y, g E(y) /p; var(y) (-p)/p egatve Bnomal Dstrbuton: y # of trals untl havng r heads y r fn( y) p ( p) r y r E(y) r/p; var(y) r(-p)/p, y r, r+, LDV-39

() Basc Posson Regresson Model Assume y d wth Posson(µ ), where µ ep( β) EX: y # of vstng doctors β µ y e β y e ( µ ) e ( e ) f( y ) y! y! ln ( ) β f y e + y ( β ) ln( y!) ( ) { ( ) β l β y β e ln( y!) } l ( β ) β Σ ( Σ y ) he Posson ML estmator of β can be vewed as a GMM estmator based on ( ) E ( y ) 0 hs moment condton s vald as long as E( y ) It means that the Posson ML estmator of β s consstent even f the Posson assumpton s ncorrect ( ) ( B β y β Σ ) ; H ( β ) Σ e If the Posson assumpton s truly correct, use [ ( ˆ )] H βpoi ML or [ B ( ˆ β )] POI ML as an estmate of Cov( ˆ β ) POI ML LDV-40

If you are not sure, use [ H ( ˆ β )] POI ML ( ˆ B βpoi ML) [ ( ˆ H βpoi ML)] For the measures of goodness of ft, see Greene In fact, we can estmat by LLS appled to y ε + wth heteroskedastc error terms Dgresson to LLS wth Heteroskedastc Errors: y h(,β) + ε Let H(,β) h (, β ) β he LLS estmator of β ( ˆL β ) mnmzes: ( (, )) Σ y h β ( ) Cov( ˆ β ) Σ H (, ˆ β ) H (, ˆ β ) where L L L Σ t L L e H(, ˆ β ) H(, ˆ β ) ( ˆ ˆ t H( t, βl) H( t, βl) ) Σ End of Dgresson e y h(, ˆ β ) ML LDV-4

() Compound Posson Model (egatve Bnomal Model): Hausman, Hall, Grlches (HHG, ECO, 984), and Cameron and rved (C, JAE, 986) β α α Assume that the y follow Posson(λ ), whereλ e µ e, µ, Ee ( α ), and the e α follow a Gamma dstrbuton: fgamma ( η) θ ( θ ) θ θη θ e η, Γ + where 0 < η < and θ t Γ ( θ ) t e dt 0 Here, α s an unobservable ndvdual effect Dgresson to Gamma dstrbuton: he most general form of the Gamma densty functon s gven: f y y e ( αβ ) α y / β gen gamma( αβ, ), Γ α where y s a contnuous postve random varable ( y > 0) E(y) αβ and var(y) αβ f ( η) s obtaned by settng y η, α θ and β /θ gamma Choose α θ and β /θ to make E(η) [a normalzaton] End of Dgresson LDV-4

ote β+ α β e β+ α y e u β y e ( e ) e ( e u ) f( y, u) y! y! hen, where f ( y ) f( y, u ) f ( u ) du gamma 0 r Γ ( θ + y ) θ y r ( r), Γ ( y + ) Γ( θ ) θ, Γ () ( ) ( ) + θ s s Γ s, and Γ () s ( s )! f s s an nteger hus, when θ s a postve nteger, ( θ + y ) f y r r θ θ θ ( ) ( ) θ + y ( ), hs s the form of the negatve bnomal dstrbuton Compound Posson egatve bnomal dstrbuton! Snce (θ+y ) follows eg-bn, E(θ+y ) θ/(-r ) e β var(y ) var(θ+y ) θ(-r )/r + θ E(y ) If we allow θ to vary over and set θ α E(y ) ; var(y ) (+α) e β + e θ β, we have hs model s called eg-bn model (HHG) LDV-43

If we set θ /α, E(y ) ; var(y ) ( ) ( ) e β αe β e β α e β + + hs model s called eg-bn model (HHG) Comment: Posson, eg-bn and eg-bn assume that E(y ) ep( β) If ths mean specfcaton s correct, the Posson, eg-bn, eg-bn MLE are all consstent as long as the true dstrbuton belongs to the lnear eponental famly [see Gourerou, Monfort and rognon (ECO, 984)] (3) estng Posson: H o : E(y ) var(y ) H a : E(y ) (Posson); β β, but var(y ) e α ( e ) s + and α 0 [If s, H a eg-bn If s, H a eg-bn ] For gven s, under H o, ( { }) E {( y µ ) y} E ( y µ ) y 0; µ µ var {( y µ ) y}, µ LDV-44

where µ hen, under H o, CL mples: L {( y ) y} Σ µ µ ( s ) Σ ( µ ) s µ (0,) Snce µ s unobservable, we need to use ˆ ˆ µ, where ˆ β s the Posson ML estmator But, we stll can show that under H o, ˆ L {( y ˆ ) y} Σ ˆ µ µ ( s ) Σ ( ˆ µ ) s ˆ µ (0,) H o may hold even f the y do not follow Posson For such cases, use ˆ L {( y ˆ ) y} Σ ˆ µ µ s ˆ µ { y} s Σ ( ˆ µ ) Σ ( y ˆ µ ) ( ) ˆ µ LDV-45

(4) Posson Model for Panel Data Assume y t d wth Posson(λ t ), where λ t ep( t β+α ) µ t ep(α ) Fed Effects Model reat α as parameters Surprsngly, MLE s consstent! λt yt e ( λt ) f( yt t ) y! t ln f ( y ) λ + y ( β + α ) ln( y!) t t t t t t (,,,, ) { ( ) t β α l βαα α t yt + t β α e ln( yt!)} l ( β, α,, α ) α t { yt e α µ t} Σ y t ln µ t t α j Σt j Σ Σ + Σ 0 Substtute these solutons nto l : l ( β ) Σ Σ y ln p ( β ), c t t t where p t ep( t β ) Σ ep( β ) t t c he MLE estmator of β based on l ( β ) s consstent even f the true dstrbuton of y t s not Posson as long long as E(y t t,α ) ep( t β)ep(α ) [Wooldrdge (JEC, 999)] For correct LDV-46

covarance matr of the ML estmator of β, use the robust form [( H ) B ( H ) ] Random Effects Model Assume that the e α follow a Gamma Dstrbuton f( y, y,, u ) f( y, α ), t t t µ tu y e ( µ tu ) t yt! yt ( µ ) ( ) t Γ θ +Σ t t yt Σ ( ( )!) ( ) t Γ θ y t t Σt µ t gamma 0 f ( y,, y, ) f( y,, y,,, u ) f ( u ) du where Q θ θ +Σ µ θ Q ( Q) y t t l βθ f y y t (, ) Σ (,,,, ) t Σ Can use the Hausman test to determne whether RE or FE s correct t y t LDV-47