S620 - Introduction To Statistical Theory - Homework 4. Enrique Areyan February 13, 2014

Σχετικά έγγραφα
ST5224: Advanced Statistical Theory II

Solution Series 9. i=1 x i and i=1 x i.

Other Test Constructions: Likelihood Ratio & Bayes Tests

2 Composition. Invertible Mappings

MATH423 String Theory Solutions 4. = 0 τ = f(s). (1) dτ ds = dxµ dτ f (s) (2) dτ 2 [f (s)] 2 + dxµ. dτ f (s) (3)

Homework 3 Solutions

CHAPTER 25 SOLVING EQUATIONS BY ITERATIVE METHODS

HOMEWORK 4 = G. In order to plot the stress versus the stretch we define a normalized stretch:

derivation of the Laplacian from rectangular to spherical coordinates

4.6 Autoregressive Moving Average Model ARMA(1,1)

Areas and Lengths in Polar Coordinates

Math221: HW# 1 solutions

C.S. 430 Assignment 6, Sample Solutions

Section 8.3 Trigonometric Equations

Areas and Lengths in Polar Coordinates

6.3 Forecasting ARMA processes

EE512: Error Control Coding

Bayesian statistics. DS GA 1002 Probability and Statistics for Data Science.

Matrices and Determinants

Problem Set 3: Solutions

Statistical Inference I Locally most powerful tests

A Two-Sided Laplace Inversion Algorithm with Computable Error Bounds and Its Applications in Financial Engineering

Section 7.6 Double and Half Angle Formulas

Απόκριση σε Μοναδιαία Ωστική Δύναμη (Unit Impulse) Απόκριση σε Δυνάμεις Αυθαίρετα Μεταβαλλόμενες με το Χρόνο. Απόστολος Σ.

Finite Field Problems: Solutions

ANSWERSHEET (TOPIC = DIFFERENTIAL CALCULUS) COLLECTION #2. h 0 h h 0 h h 0 ( ) g k = g 0 + g 1 + g g 2009 =?

Space-Time Symmetries

3.4 SUM AND DIFFERENCE FORMULAS. NOTE: cos(α+β) cos α + cos β cos(α-β) cos α -cos β

DESIGN OF MACHINERY SOLUTION MANUAL h in h 4 0.

Econ 2110: Fall 2008 Suggested Solutions to Problem Set 8 questions or comments to Dan Fetter 1

( ) 2 and compare to M.

6.1. Dirac Equation. Hamiltonian. Dirac Eq.

An Introduction to Signal Detection and Estimation - Second Edition Chapter II: Selected Solutions

Practice Exam 2. Conceptual Questions. 1. State a Basic identity and then verify it. (a) Identity: Solution: One identity is csc(θ) = 1

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 19/5/2007

Example Sheet 3 Solutions

Solutions to Exercise Sheet 5

The Simply Typed Lambda Calculus

Partial Differential Equations in Biology The boundary element method. March 26, 2013

Quadratic Expressions

Section 9.2 Polar Equations and Graphs

Inverse trigonometric functions & General Solution of Trigonometric Equations

Similarly, we may define hyperbolic functions cosh α and sinh α from the unit hyperbola

Phys460.nb Solution for the t-dependent Schrodinger s equation How did we find the solution? (not required)

2. THEORY OF EQUATIONS. PREVIOUS EAMCET Bits.

Concrete Mathematics Exercises from 30 September 2016

SOLUTIONS TO MATH38181 EXTREME VALUES AND FINANCIAL RISK EXAM

A Note on Intuitionistic Fuzzy. Equivalence Relation

w o = R 1 p. (1) R = p =. = 1

CHAPTER 101 FOURIER SERIES FOR PERIODIC FUNCTIONS OF PERIOD

Uniform Convergence of Fourier Series Michael Taylor

F19MC2 Solutions 9 Complex Analysis

Homework 8 Model Solution Section

The challenges of non-stable predicates

Statistics 104: Quantitative Methods for Economics Formula and Theorem Review

Theorem 8 Let φ be the most powerful size α test of H

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 6/5/2006

Exercises 10. Find a fundamental matrix of the given system of equations. Also find the fundamental matrix Φ(t) satisfying Φ(0) = I. 1.

k A = [k, k]( )[a 1, a 2 ] = [ka 1,ka 2 ] 4For the division of two intervals of confidence in R +

forms This gives Remark 1. How to remember the above formulas: Substituting these into the equation we obtain with

5.4 The Poisson Distribution.

SCHOOL OF MATHEMATICAL SCIENCES G11LMA Linear Mathematics Examination Solutions

Variational Wavefunction for the Helium Atom

b. Use the parametrization from (a) to compute the area of S a as S a ds. Be sure to substitute for ds!

Lecture 2. Soundness and completeness of propositional logic

Second Order Partial Differential Equations

Srednicki Chapter 55

Jesse Maassen and Mark Lundstrom Purdue University November 25, 2013

Numerical Analysis FMN011

Nowhere-zero flows Let be a digraph, Abelian group. A Γ-circulation in is a mapping : such that, where, and : tail in X, head in

Fourier Series. MATH 211, Calculus II. J. Robert Buchanan. Spring Department of Mathematics

Every set of first-order formulas is equivalent to an independent set

Math 446 Homework 3 Solutions. (1). (i): Reverse triangle inequality for metrics: Let (X, d) be a metric space and let x, y, z X.

Parametrized Surfaces

Ordinal Arithmetic: Addition, Multiplication, Exponentiation and Limit

Exercises to Statistics of Material Fatigue No. 5

( )( ) ( ) ( )( ) ( )( ) β = Chapter 5 Exercise Problems EX α So 49 β 199 EX EX EX5.4 EX5.5. (a)

Approximation of distance between locations on earth given by latitude and longitude

CRASH COURSE IN PRECALCULUS

Overview. Transition Semantics. Configurations and the transition relation. Executions and computation

Second Order RLC Filters

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 24/3/2007

Tridiagonal matrices. Gérard MEURANT. October, 2008

SOLVING CUBICS AND QUARTICS BY RADICALS

6. MAXIMUM LIKELIHOOD ESTIMATION

STAT200C: Hypothesis Testing

PARTIAL NOTES for 6.1 Trigonometric Identities

Potential Dividers. 46 minutes. 46 marks. Page 1 of 11

ECE598: Information-theoretic methods in high-dimensional statistics Spring 2016

Fractional Colorings and Zykov Products of graphs

Math 6 SL Probability Distributions Practice Test Mark Scheme

Chapter 6: Systems of Linear Differential. be continuous functions on the interval

ENGR 691/692 Section 66 (Fall 06): Machine Learning Assigned: August 30 Homework 1: Bayesian Decision Theory (solutions) Due: September 13

Pg The perimeter is P = 3x The area of a triangle is. where b is the base, h is the height. In our case b = x, then the area is

Μηχανική Μάθηση Hypothesis Testing

Reminders: linear functions

= λ 1 1 e. = λ 1 =12. has the properties e 1. e 3,V(Y

ORDINAL ARITHMETIC JULIAN J. SCHLÖDER

Strain gauge and rosettes

Aquinas College. Edexcel Mathematical formulae and statistics tables DO NOT WRITE ON THIS BOOKLET

Transcript:

S6 - Introuction To Statistical Theory - Homework 4 (3.) Let θ be a ranom variable in (, ) with ensity where β, γ (, ) (i) Calculate the mean an moe of θ. Enrique Areyan February 3, 4 π(θ) θ γ e βθ, Solution: Since θ is proportional to θ γ e βθ, we know that θ Gamma(γ, β), where β >, γ >. Therefore, E[θ] γ γ, an Moe β β (ii) Suppose that X,..., X n are ranom variables, which, conitional on θ are inepenent an each have the Poisson istribution with parameter θ. Fin the form of the posterior ensity of θ given observe values X x,... X n x n. What is the posterior mean? Solution: We know that x θ P oisson(θ). Let us compute the posterior istribution: θ x π(θ)f(x; θ) Θ π(θ )f(x; θ )θ by efinition of posterior ensity θ γ e βθ e θ θ x constant x! constant replacing prior an Poisson istribution constant θ γ +x e βθ θ grouping like terms constant θ (γ+x) e (β+)θ rearranging exponents Gamma(γ + x, β + ), by efinition of Gamma istribution Therefore, the posterior mean is the mean of the gamma istribution with parameters γ + x an β +, i.e., E[θ x] γ + x β + Note that γ > an x,,,, so γ + x > an β > thus β + >. (ii) Suppose now that T,..., T n are ranom variables, which, conitional on θ are inepenent, each exponentially istribute with parameter θ. What is the moe of the posterior istribution of θ, given T t,..., T n t n? Solution: In this case we know that x θ Exp(θ). The posterior istribution is: θ x π(θ)f(x; θ) Θ π(θ )f(x; θ )θ by efinition of posterior ensity θ γ e βθ constant θe θt constant replacing prior an Exp. istribution constant θ γ + e βθ θt grouping like terms constant θ γ e (β+t)θ rearranging exponents Gamma(γ +, β + t), by efinition of Gamma istribution

Since γ > then γ + >, an we have a close form for the moe: Moe (γ + ) β + t γ β + t (3.3) Fin the form of the Bayes rule in an estimation problem with loss function { a(θ ) if θ, L(θ, ) b( θ) if > θ, where a an b are given positive constants. Solution: We wish to minimize the following expression: Θ L(θ, )π(θ x)θ As a function of, i.e., let f() be efine as follow: Fin f () f() a(θ )π(θ x)θ + a(θ )π(θ x)θ + b( θ)π(θ x)θ [ ] [ a θπ(θ x)θ π(θ x)θ + b [ b b( θ)π(θ x)θ π(θ x)θ ] π(θ x)θ a π(θ x)θ + a θπ(θ x)θ b θπ(θ x)θ θπ(θ x)θ (erivative of f with respect to ) using prouct rule an funamental theorem of calculus: f () ( b ) π(θ x)θ a π(θ x)θ + [bπ( x) + aπ( x)] aπ( x) bπ( x) ] b π(θ x)θ a b π(θ x)θ a π(θ x)θ + (a + b)π( x) (a + b)π( x) π(θ x)θ Set f () b π(θ x)θ a π(θ x)θ b π(θ x)θ a π(θ x)θ b a This expression represents a quantile of posterior π(θ x) in terms of a an b. Concretely: π(θ x)θ π(θ x)θ π(θ x)θ + π(θ x)θ since prob. as to π(θ x)θ + b a π(θ x)θ replacing for our minimun π(θ x)θ a a + b Solving for the posterior quantile

Hence, the Bayes rule is the quantile posterior a a + b of π(θ x). Note that we shoul check whether the minimality of the value obtain. Using the secon erivative test: f () b π(θ x)θ a π(θ x)θ bπ( x) + aπ( x) (a + b)π( x) (a + b) () The last equality is hols since the probability of taking action (x) given x is. By hypothesis a an b are given positive constants, an so is a + b, so we have a local minimun. Moreover, this expression is always positive an hence, the minimum foun is the global minimum. (3.4) Suppose that X is istribute as a binomial ranom variable with inex n an parameter θ. Calculate the Bayes rule (base on the single observation X) for estimating θ when the prior istribution is the uniform istribution on [, ] an the loss function is Is the rule you obtain minimax? L(θ, ) (θ ) /{θ( θ)} Solution: We wish to minimize the following expression: Θ L(θ, )π(θ x)θ Let us first fin the posterior istribution π(θ x). (θ ) θ( θ) π(θ x)θ θ x π(θ)f(x; θ) Θ π(θ )f(x; θ )θ by efinition of posterior ensity (n) x θ x ( θ) n x replacing prior an Binomial istribution Θ π(θ )f(x; θ )θ constant θ x ( θ) n x grouping like terms Beta(x +, n x + ), by efinition of Beta istribution Now we can minimize the function: (θ ) f() π(θ x)θ efine f() this way θ( θ) (θ ) θ x ( θ) n x θ( θ) B(x +, n x + ) θ replace posterior B(x +, n x + ) (θ ) θ x ( θ) n x θ simplifying Note that in this context is a constant, an so we can ignore this term for optimization B(x +, n x + ) purposes. Hence, optimize f() is equivalent to optimize g() efine as: g() (θ ) θ x ( θ) n x θ (θ θ + )θ x ( θ) n x θ θ x+ ( θ) n x θ θ x ( θ) n x θ + θ x ( θ) n x θ

Fin g () g () (erivative of g with respect to ): [ () θ x+ ( θ) n x θ θ x ( θ) n x θ + θ x ( θ) n x θ + θ x ( θ) n x θ ] θ x ( θ) n x θ Set g () θ x ( θ) n x θ + θ x ( θ) n x θ We can write this conition in terms of the Beta function, i.e., B(a, b) B(x +, n x) B(x, n x) θ x ( θ) n x θ θ x ( θ) n x θ t a ( t) b t, as follows: Finally, using ientities relating the Beta an Gamma function an the fact that n an x are positive integers: B(x +, n x) B(x, n x) Γ(x + )Γ(n x) Γ(n + ) Γ(x)Γ(n x) Γ(n) Γ(x + )Γ(n) x!(n )! Γ(n + )Γ(x) n!(x )! x n An hence, the Bayes rule is just the mean π X. A similar argument to that in (3.3), using secon n erivative test, shows that this is in fact the minimum. The obtaine rule π is minimax since it has constant risk, as the following computation shows: R(θ, π ) E θ L(θ, π ) efinition of risk ( E θ L θ, X ) n replacing for the rule π ( θ X n E θ θ( θ) ) by loss function θ θ X ( X n + n E θ θ( θ) ) squaring θ θ( θn n ) + nθ( θ)+n θ n θ( θ) expectation of a Binomial an Binomial square θ + nθ( θ) + n θ n θ( θ) nθ nθ + n θ n θ n θ( θ)

nθ nθ n θ( θ) n( θ) n ( θ) n Since n is fixe, this is a constant. (3.5) At a critical stage in the evelopment of a new airplane, a ecision must be taken to continue or to abanon the project. The financial viability of the project can be measure by a parameter θ, < θ <, the project being profitable if θ >. Data x provie information about θ. (i) If θ <, the cost to the taxpayer of continuing the project is ( θ) (in units of $bilion), whereas if θ > it is zero (since the project will be privatize if profitable). If θ > the cost of abanoning the project is (θ ) (ue to contractual arrangements for purchasing the airplane from the French), whereas if θ < it is zero. Derive the Bayes ecision rule in terms of the posterior mean of θ given x. Solution: Let A {, }, where enotes abanoning the project an continuing the project. Let us write the loss function as follow: L(θ, ) We want to minimize Θ L(θ, )π(θ x)θ, i.e., { { if θ ( (θ ) if < θ L(θ, ) θ) if θ if < θ Θ L(θ, )π(θ x)θ ( θ)π(θ x)θ if (θ )π(θ x)θ if The Bayes rule is that which minimizes the above expression. That is, the Bayes rule will choose to go on with the protect if ( θ)π(θ x)θ < (θ )π(θ x)θ or, equivalently, the Bayes rule will choose to abanon the protect if (θ )π(θ x)θ < ( θ)π(θ x)θ. (ii) The minister of aviation has prior ensity 6θ( θ) for θ. The Prime Minister has prior ensity 4θ 3. The prototype aeroplane is subjecte to trials, each inepenently having probability θ of success, an the ata x consist of the total number of trials require for the first successful result to be obtaine. For what values of x will there be serious ministerial isagreement? Solution: Clearly, X Geometric(θ), where X,,.... Hence π(x) ( θ) x θ. We can compute the posterior probability ensity π(θ x) for the minister of aviation (MA) an the Prime Minister (PM ): MA: π(θ x) constant θ( θ)( θ) x θ θ ( θ) x Beta(3, x + ) PM π(θ x) constant θ 3 ( θ) x θ θ 4 ( θ) x Beta(5, x) We want to fin the values of x for which the Bayes rule for MA an MP will isagree. Let π MA (θ x) an π MP (θ x) be the prior of the minister of aviation an Prime Minister respectively. Then, the values of x we are intereste in are:

π MA (θ x)θ π MP (θ x)θ x for which: ( θ)π MA(θ x)θ < ( θ)π MP (θ x)θ > (θ )π MA(θ x)θ (θ )π MP (θ x)θ Solving for x using the fact that π(θ x) is a probability istribution an computing its mean: Note that θπ MA (θ x)θ < θπ MA (θ x)θ π MA (θ x)θ θπ MP (θ x)θ > θπ MP (θ x)θ π MP (θ x)θ π MA (θ x)θ θπ MA (θ x)θ < π MP (θ x)θ θπ MP (θ x)θ > π MA (θ x)θ an θπ MA (θ x)θ E[π MA (θ x)]. Same for MP. We know each of these istributions are Beta an so we know its mean: E[π MA (θ x)] E[Beta(3, x + )] 3 x + 4 an E[π MP (θ x)] E[Beta(5, x)] 5 5 + x Replacing into our conitions: E[π MA(θ x)] < 3 x + 4 < E[π MP (θ x)] > 5 5 + x > < 3 x + 4 > 5 5 + x x + 4 < 6 x + 5 > x < x > 5 Hence, the minister of aviation will go on with the project if x < while the Prime Minister will choose to abanon the project if x > 5. This means that MA will choose to abanon if x >. So, for values x 3, 4; there will be a isagreement: MA woul want to abanon but PM will not. A ecision will have to be reache through some other argument. Note that this result makes intuitive sense: the prime minister prior is a cubic function that places lower emphasis on smaller values of θ an thus, is more tolerant to risk while the opposite is try for the minister of aviation whose prior is a quaratic function.