Debabrata Dey and Atanu Lahiri

Similar documents
7.0 Equality Contraints: Lagrange Multipliers

Some Different Perspectives on Linear Least Squares

Chapter 9 Jordan Block Matrices

arxiv:math/ v1 [math.gm] 8 Dec 2005

Non-degenerate Perturbation Theory

Solutions to problem set ); (, ) (

A Penalty Function Algorithm with Objective Parameters and Constraint Penalty Parameter for Multi-Objective Programming

PTAS for Bin-Packing

Applying the condition for equilibrium to this equilibrium, we get (1) n i i =, r G and 5 i

for each of its columns. A quick calculation will verify that: thus m < dim(v). Then a basis of V with respect to which T has the form: A

Solutions for HW4. x k n+1. k! n(n + 1) (n + k 1) =.

5 Short Proofs of Simplified Stirling s Approximation

An Expansion of the Derivation of the Spline Smoothing Theory Alan Kaylor Cline

Sebastián Martín Ruiz. Applications of Smarandache Function, and Prime and Coprime Functions

Part 4b Asymptotic Results for MRR2 using PRESS. Recall that the PRESS statistic is a special type of cross validation procedure (see Allen (1971))

2/20/2013. Topics. Power Flow Part 1 Text: Power Transmission. Power Transmission. Power Transmission. Power Transmission

18.413: Error Correcting Codes Lab March 2, Lecture 8

A Conventional Approach for the Solution of the Fifth Order Boundary Value Problems Using Sixth Degree Spline Functions

Algorithms behind the Correlation Setting Window

ANALYSIS ON THE NATURE OF THE BASIC EQUATIONS IN SYNERGETIC INTER-REPRESENTATION NETWORK

Mu Sequences/Series Solutions National Convention 2014

Standard Deviation for PDG Mass Data

THE TRUNCATED RANDIĆ-TYPE INDICES

Lecture 3 Probability review (cont d)

MATH 247/Winter Notes on the adjoint and on normal operators.

CIS 800/002 The Algorithmic Foundations of Data Privacy October 13, Lecture 9. Database Update Algorithms: Multiplicative Weights

TESTS BASED ON MAXIMUM LIKELIHOOD

Econometric Methods. Review of Estimation

{ }{ ( )} (, ) = ( ) ( ) ( ) Chapter 14 Exercises in Sampling Theory. Exercise 1 (Simple random sampling): Solution:

2. Independence and Bernoulli Trials

å 1 13 Practice Final Examination Solutions - = CS109 Dec 5, 2018

Discrete Mathematics and Probability Theory Fall 2016 Seshia and Walrand DIS 10b

( ) ( ) ( ( )) ( ) ( ) ( ) ( ) ( ) = ( ) ( ) + ( ) ( ) = ( ( )) ( ) + ( ( )) ( ) Review. Second Derivatives for f : y R. Let A be an m n matrix.

Assignment 5/MATH 247/Winter Due: Friday, February 19 in class (!) (answers will be posted right after class)

2SLS Estimates ECON In this case, begin with the assumption that E[ i

Chapter 4 (Part 1): Non-Parametric Classification (Sections ) Pattern Classification 4.3) Announcements

A tighter lower bound on the circuit size of the hardest Boolean functions

i 2 σ ) i = 1,2,...,n , and = 3.01 = 4.01

F. Inequalities. HKAL Pure Mathematics. 進佳數學團隊 Dr. Herbert Lam 林康榮博士. [Solution] Example Basic properties

MA 524 Homework 6 Solutions

Polyphase Filters. Section 12.4 Porat

12.2 Estimating Model parameters Assumptions: ox and y are related according to the simple linear regression model

Introduction to local (nonparametric) density estimation. methods

UNIT 2 SOLUTION OF ALGEBRAIC AND TRANSCENDENTAL EQUATIONS

CHAPTER 4 RADICAL EXPRESSIONS

Third handout: On the Gini Index

Camera calibration & radiometry

Lecture 9: Tolerant Testing

Some results and conjectures about recurrence relations for certain sequences of binomial sums.

Bounds on the expected entropy and KL-divergence of sampled multinomial distributions. Brandon C. Roy

Lecture 07: Poles and Zeros

A Family of Non-Self Maps Satisfying i -Contractive Condition and Having Unique Common Fixed Point in Metrically Convex Spaces *

9 U-STATISTICS. Eh =(m!) 1 Eh(X (1),..., X (m ) ) i.i.d

Chapter 5 Properties of a Random Sample

= lim. (x 1 x 2... x n ) 1 n. = log. x i. = M, n

2.28 The Wall Street Journal is probably referring to the average number of cubes used per glass measured for some population that they have chosen.

STK4011 and STK9011 Autumn 2016

Non-uniform Turán-type problems

Decomposition of Hadamard Matrices

KURODA S METHOD FOR CONSTRUCTING CONSISTENT INPUT-OUTPUT DATA SETS. Peter J. Wilcoxen. Impact Research Centre, University of Melbourne.

Algorithms Theory, Solution for Assignment 2

A New Method for Solving Fuzzy Linear. Programming by Solving Linear Programming

On Eccentricity Sum Eigenvalue and Eccentricity Sum Energy of a Graph

3.1 Introduction to Multinomial Logit and Probit

The Mathematical Appendix

MOLECULAR VIBRATIONS

Solutions to Odd-Numbered End-of-Chapter Exercises: Chapter 17

44 Chapter 3. Find the 13 term and the sum of the first 9 terms of the geometric sequence 48, 24, 12, 6, 3, 3 2 Solution 2

Sampling Theory MODULE X LECTURE - 35 TWO STAGE SAMPLING (SUB SAMPLING)

Lecture 12 APPROXIMATION OF FIRST ORDER DERIVATIVES

Multiple Choice Test. Chapter Adequacy of Models for Regression

D. L. Bricker, 2002 Dept of Mechanical & Industrial Engineering The University of Iowa. CPL/XD 12/10/2003 page 1

Arithmetic Mean and Geometric Mean

Unique Common Fixed Point of Sequences of Mappings in G-Metric Space M. Akram *, Nosheen

( ) H α iff α Pure and Impure Altruism C H,H S,T. Find the utility payoff matrix of PD if subjects all have utility u C D

1 Solution to Problem 6.40

Chapter 13, Part A Analysis of Variance and Experimental Design. Introduction to Analysis of Variance. Introduction to Analysis of Variance

THE ROYAL STATISTICAL SOCIETY 2016 EXAMINATIONS SOLUTIONS HIGHER CERTIFICATE MODULE 5

d dt d d dt dt Also recall that by Taylor series, / 2 (enables use of sin instead of cos-see p.27 of A&F) dsin

ε. Therefore, the estimate

Laboratory I.10 It All Adds Up

Strong Convergence of Weighted Averaged Approximants of Asymptotically Nonexpansive Mappings in Banach Spaces without Uniform Convexity

Jacobi symbols. p 1. Note: The Jacobi symbol does not necessarily distinguish between quadratic residues and nonresidues. That is, we could have ( a

PROJECTION PROBLEM FOR REGULAR POLYGONS

Capacitated Plant Location Problem:

The internal structure of natural numbers, one method for the definition of large prime numbers, and a factorization test

Feature Selection: Part 2. 1 Greedy Algorithms (continued from the last lecture)

Section l h l Stem=Tens. 8l Leaf=Ones. 8h l 03. 9h 58

PPCP: The Proofs. 1 Notations and Assumptions. Maxim Likhachev Computer and Information Science University of Pennsylvania Philadelphia, PA 19104

Investigation of Partially Conditional RP Model with Response Error. Ed Stanek

ENGI 4421 Propagation of Error Page 8-01

Ordinary Least Squares Regression. Simple Regression. Algebra and Assumptions.

X ε ) = 0, or equivalently, lim

Q-analogue of a Linear Transformation Preserving Log-concavity

Lecture 02: Bounding tail distributions of a random variable

Summary of the lecture in Biostatistics

b. There appears to be a positive relationship between X and Y; that is, as X increases, so does Y.

PRACTICAL CONSIDERATIONS IN HUMAN-INDUCED VIBRATION

-Pareto Optimality for Nondifferentiable Multiobjective Programming via Penalty Function

are positive, and the pair A, B is controllable. The uncertainty in is introduced to model control failures.

Transcription:

RESEARCH ARTICLE QUALITY COMPETITION AND MARKET SEGMENTATION IN THE SECURITY SOFTWARE MARKET Debabrata Dey ad Atau Lahr Mchael G. Foster School of Busess, Uersty of Washgto, Seattle, Seattle, WA 9895 U.S.A. {ddey@uw.edu} {lahra@uw.edu} Guoyg Zhag Dllard College of Busess, Mdwester State Uersty, Wchta Falls, Wchta Falls, TX 76308 U.S.A. {grace.zhag@wsu.edu} Apped A Proofs Proof of Lea Settg u, fro Fgure, we ca fd the aret coerage of θ,,,,, as u u, whch ca be sued oer to obta u G. Substtutg ths to ( for ad otg that p 0 θ 0 0, we fd p Gθu Gθ (A We wll ow proe the lea by ducto. It s clear that (3 reduces to (A for. Let (3 hold for, plyg p G θ θ We substtute ths to ( for to obta ( θ θ p G u p G( θ G θ θ θ G( θ θ G θ θ θ θ G θ θ I other words, f (3 holds for, the t also holds for. Sce (3 holds for, the proof s ow coplete. # MIS Quarterly Vol. 38 No. Apped/Jue 04 A

Dey et al./qualty Copetto & Maret Segetato Proof of Proposto ( Sce R GH c(θ ad <, we ca use the frst order codto wth respect to to obta GH GH Gθ θ 0 θ Therefore, we get GH θ θ GH G ( H H ( ( Now, fro defto of H l, l,,,, we get l l Hl Hl l l l θ θ θ θ l ( θ θ l l l Sce θ > θ ples that >, sug the aboe oer l, we get H H ( Hl Hl ( θl θl l l l (A Now, sce θ > θ, there ust est soe l, < l <, such that θ l θ l- > 0, plyg that the rght had sde of (A s strctly greater tha zero. Thus, H H > 0 ad, hece, θ θ > 0, whch copletes the proof. ( Frst, we ote that, for all,,,, Y θ G g whch, of course, eas that Y H Y θ. Net, we also ow that ( gg ( GH GG Y H gy Y H θ θ θ θ θ θ ( θ 0 Therefore, Y > H, ad hece Y > θ, for all,,,. Sug both sdes oer, we get Y θ Y Y θ Y ( If θ θ, the for eery l, < l <, θ l θ l- 0. Therefore, fro (A, H H 0 plyg θ θ 0 or. ( Fro the proof of part (, we ow that Y > θ. Therefore, R ( Y θ θ 0 θ g A MIS Quarterly Vol. 38 No. Apped/Jue 04

Dey et al./qualty Copetto & Maret Segetato O the other had, fro the proof of part (, we ow that GH θ G ; we ca the show that R g H YH θ( H Y < 0 I other words, whe g creases, the frst order respose by a edor to ths chage s to crease qualty ad decrease aret share. Howeer, whe θ, the edor caot crease qualty ay further ad ts oly frst order respose would be to decrease ts aret share. Sce such a respose copleets other edors actos, equlbru, ust decrease. ( We proe ths part by cotradcto. Let there be a equlbru wth θ θ <, for soe <, wth <, <. We ow that edor soles the followg azato proble: Ma R G θ θ c θ θ, Sce θ < (by assupto, the frst order codto wth respect to θ ust be satsfed: ( G g θ θ c ( θ θ 0 Furtherore, sce θ θ, for all, < <, we ow fro aboe that the aret shares of these edors would be equal. We set θ θ θ ad to get c ( θ G g θ g θ G g θ g θ θ G ( g θ ( g θ ( θ G ( g θ g θ (A3 We ow cosder how the reeue of the th edor chages wth the qualty of ts ow product: G g θ g θ c ( θ θ Substtutg (A3 to the aboe epresso, we get G ( < 0, whch s a olato of the frst order codto for a teror θ soluto. # Proof of Theore We frst show that there ests a g beyod whch all edors offer a qualty leel of oe. To see ths, cosder edor. Its proft s ge by R GH c(θ. Therefore, Gg c θ ( θ ( θ MIS Quarterly Vol. 38 No. Apped/Jue 04 A3

Dey et al./qualty Copetto & Maret Segetato Fro (4, Y Σ θ ad G gy. Furtherore, fro the proof of Proposto (, we ow that Therefore, we hae GH θ G equlbru. ( gy ( Y G gθ ( gy gy The last equalty results fro the fact that H < Y; see the proof of Proposto (. Now, fro that proof, we also ow that Y ( gy g gy. Furtherore, s a creasg fucto of gy. Hece, we ca wrte gy Gg θ g ( gy ( ( g gy g ( ( g whch s clearly a creasg fucto of g. Sce cn( s bouded, for a suffcetly large g, we wll hae θ θ ( g ( ( g ( c > 0 Sce c (@ s a creasg coe fucto, the aboe eas that, equlbru, a teror soluto s ot possble ad θ. Ths, tur, ples that θ, for all,,. I other words, there ust est a threshold for g we characterze ths threshold as γ (c Theore beyod whch ertcal dfferetato would dsappear. We ow cosder what happes whe g starts decreasg below ths threshold. Of course, f the deelopet cost s eglgble, trally, all edors would cotue to offer a qualty leel of oe, rrespecte of the alue of g. Howeer, f the deelopet cost s sgfcat, soe edors would hae to drop ther qualty leel below oe, but we wll show that they ca do so oly oe edor at a te. To proe ths last cla, suppose that two edors drop the qualty leel to below oe at the sae te. At the alue of g where ths occurs, these edors ust be barely at the sae teror soluto. Howeer, fro the proof of Proposto (, t s clear that o two edors ca hae the sae teror soluto. Therefore, whe g decreases, edors would ot oly drop ther qualty leels fro oe, but would also do so oly oe at a te, whle atag the order of ther qualty leels. Equaletly, as g creases, ther qualtes would reach oe at dfferet alues of g. It s also clear fro the proof of Proposto ( that, oce a qualty leel reaches oe, t caot drop whe g creases further. Tae together, t s clear that, as g creases, the segetato leel the aret gradually decreases. #, so Proof of Theore γ To proe the estece of the erse fucto, t s suffcet to show that γ (g s a strctly ootoc fucto. It turs out that g g. To see ths, we obsere that g where ( ( γ ( g ( ( A B g 4g g ( ( ( 3 3 3 4 3 3 A 8g 8g 6 3g 0g 6 8g g 4g 3g, B C 4g g, ad C 4g 4 5g 3g g 3 ( > 0 3 3 Now A B 4g g D, where D 4g g ; hece, A > B, or A > B, as log as D > 0. If g <, D s always poste. ( g Therefore, we oly cosder the case where g >. I that case, D > 0 f ad oly f <. Suppose ot. The, there s a δ > 0 such that δ ( g. Substtutg ths to C leads to g g A4 MIS Quarterly Vol. 38 No. Apped/Jue 04

Dey et al./qualty Copetto & Maret Segetato ( g ( g ( g 3 C 3 δ 3 ( 8 ( gg 9g g6 g ( 3g g The frst ad the thrd ters are clearly egate sce g >. Furtherore, sce δ > 0, whe g >, t ca be show, after soe algebra, that the secod ter caot be poste. Therefore, C < 0, plyg B < 0. Sce A > 0 always, ths, tur, ples that A > B, whch copletes the proof of the frst part. For the secod part, we ote that the olgopoly equlbru ca be oly oe of ( regos. Let Rego I deote the rage of g alues wth the frst aret cofgurato, where all edors offer the qualty leel of oe. Slarly, let Rego II be the rage for the secod oe, where oly the lowest qualty edor, aely edor, offers a qualty leel below oe (θ <. Vertcal dfferetato wll be obsered as soo as the equlbru outcoe oes out of Rego I. Therefore, we oly eed to eae the boudary betwee Regos I ad II. I both the regos, θ θ 3 θ, ad t follows fro Proposto ( that 3. Let h deote ths coo aret share. The optzato proble of edor ca, therefore, be splfed to ( ( h ( ( h ( δ 3 ( ( ; Ma R θ g θ c θ θ, s.t. θ > 0, h The followg frst order codto ust be satsfed by the soluto of the ucostraed proble: ( θ θ( ( ( θ ( ( h g( ( h g h c 0 θ Sce, at the boudary of Regos I ad II, θ, we substtute t aboe to obta ( ( ( ( ( ( c g h h γ (A4 Now, whe θ s are all oe, frst order codtos wth respect to,,,,, result whch ca be substtuted to (A4 to obta where h ( g( 4g( g ( g( γ ( g ( g ( g ( 3 μ g 3 ( ( ( ( ( 6 4 ( ( 3 5 4 μ g g g g ad ( 3 4 ( ( ( ( ( g g g g g Therefore, the codto cn( > γ (g whch s equalet to g < γ (cn( esures that the outcoe s ot Rego I. # Proof of Proposto Sce γ ( g ( g ( g ( 3 μ g The result follows drectly fro the aboe. #, usg l Hosptal s rule twce, we get γ ( 0 lγ ( g ( 0 ( 0 ( μ g 4 0 3 MIS Quarterly Vol. 38 No. Apped/Jue 04 A5

Dey et al./qualty Copetto & Maret Segetato Proof of Theore 3 Recall that where, as before Therefore, we hae Cobg the aboe, we ca wrte ad The aboe epressos lead to R GH GH G Hl l c( θ t t t G gy, Y θ, ad H θ θ GH G H H G G θ gθ ad H f θ otherwse G H H G G H lt l H t lt ( GH Gθ gθ H Gθ gθ H Gθ gθh t lt lt lt t G H H G GH G H H G G H lt l H t lt ( GH Gθ gθ H Gθ gθ H Gθ gθ H t lt lt lt t G G θ θ θθ θ θ R R R G ( θh θh ( θ θ θ θl l θθ t t t (A5 We ow obsere θh θh θθ θθ θθ θθ ( θθ θ θ θ θ θ ( θ θ θ θ θ θ ( θ θ θ θ ( θ θ θ Substtutg ths to (A5 leads to A6 MIS Quarterly Vol. 38 No. Apped/Jue 04

Dey et al./qualty Copetto & Maret Segetato G ( ( θ θ θθ θ θ θ θ θ θ θ l l t t t Sce θ s the largest aog all the ersos proded, t s easy to see that the rght had sde of the aboe epresso s poste, whch s a olato of the codto (7. # Proof of Lea Ths proof s slar to that of Lea. We ca show that (9 ples p gθ θ θ 0 0 Substtutg 0 Σ ad θ 0 φ ad rearragg ters, we get (0. # Proof of Theore 4 It s slar to the proofs of Theores ad, wth gn g( φ. # MIS Quarterly Vol. 38 No. Apped/Jue 04 A7