e-companion ONLY AVAILABLE IN ELECTRONIC FORM

Similar documents
2.6 Rational Functions and Their Graphs

AUTOMATIC GENERATION OF FUZZY PAYOFF MATRIX IN GAME THEORY

ELEMENTARY PORTFOLIO MATHEMATICS


Monopoly vs. Competition in Light of Extraction Norms. Abstract

1 Random Variables and Key Statistics

CHAPTER 8 Estimating with Confidence

Math 312, Intro. to Real Analysis: Homework #4 Solutions

We learned: $100 cash today is preferred over $100 a year from now

A random variable is a variable whose value is a numerical outcome of a random phenomenon.

5. Best Unbiased Estimators

Lecture 9: The law of large numbers and central limit theorem

EVEN NUMBERED EXERCISES IN CHAPTER 4

MATH : EXAM 2 REVIEW. A = P 1 + AP R ) ny

Forecasting bad debt losses using clustering algorithms and Markov chains

CAPITAL ASSET PRICING MODEL

FINM6900 Finance Theory How Is Asymmetric Information Reflected in Asset Prices?

SELECTING THE NUMBER OF CHANGE-POINTS IN SEGMENTED LINE REGRESSION

Dr. Maddah ENMG 624 Financial Eng g I 03/22/06. Chapter 6 Mean-Variance Portfolio Theory

CHAPTER 2 PRICING OF BONDS

Rafa l Kulik and Marc Raimondo. University of Ottawa and University of Sydney. Supplementary material

14.30 Introduction to Statistical Methods in Economics Spring 2009

Supersedes: 1.3 This procedure assumes that the minimal conditions for applying ISO 3301:1975 have been met, but additional criteria can be used.

Maximum Empirical Likelihood Estimation (MELE)

Models of Asset Pricing

Combining imperfect data, and an introduction to data assimilation Ross Bannister, NCEO, September 2010

Statistics for Economics & Business

Notes on Expected Revenue from Auctions

Estimating Volatilities and Correlations. Following Options, Futures, and Other Derivatives, 5th edition by John C. Hull. Chapter 17. m 2 2.

INTERVAL GAMES. and player 2 selects 1, then player 2 would give player 1 a payoff of, 1) = 0.

Models of Asset Pricing

Models of Asset Pricing

1 Estimating sensitivities

An Introduction to Certificates of Deposit, Bonds, Yield to Maturity, Accrued Interest, and Duration

Combinatorial Proofs of Fibonomial Identities

Sequences and Series

11.7 (TAYLOR SERIES) NAME: SOLUTIONS 31 July 2018

Inferential Statistics and Probability a Holistic Approach. Inference Process. Inference Process. Chapter 8 Slides. Maurice Geraghty,

Optimizing of the Investment Structure of the Telecommunication Sector Company

SUPPLEMENTAL MATERIAL

A New Constructive Proof of Graham's Theorem and More New Classes of Functionally Complete Functions

Sampling Distributions and Estimation

Online appendices from Counterparty Risk and Credit Value Adjustment a continuing challenge for global financial markets by Jon Gregory

Introduction to Statistical Inference

EXERCISE - BINOMIAL THEOREM

point estimator a random variable (like P or X) whose values are used to estimate a population parameter

Chapter 8. Confidence Interval Estimation. Copyright 2015, 2012, 2009 Pearson Education, Inc. Chapter 8, Slide 1

T4032-BC, Payroll Deductions Tables CPP, EI, and income tax deductions British Columbia Effective January 1, 2016

. The firm makes different types of furniture. Let x ( x1,..., x n. If the firm produces nothing it rents out the entire space and so has a profit of

Appendix 1 to Chapter 5

Exam 2. Instructor: Cynthia Rudin TA: Dimitrios Bisias. October 25, 2011

BASIC STATISTICS ECOE 1323

Problem Set 1a - Oligopoly

0.1 Valuation Formula:

of Asset Pricing R e = expected return

Overlapping Generations

First determine the payments under the payment system

1. Suppose X is a variable that follows the normal distribution with known standard deviation σ = 0.3 but unknown mean µ.

Parametric Density Estimation: Maximum Likelihood Estimation

T4032-MB, Payroll Deductions Tables CPP, EI, and income tax deductions Manitoba Effective January 1, 2016

T4032-ON, Payroll Deductions Tables CPP, EI, and income tax deductions Ontario Effective January 1, 2016

5 Statistical Inference

Institute of Actuaries of India Subject CT5 General Insurance, Life and Health Contingencies

Estimating Proportions with Confidence

APPLICATION OF GEOMETRIC SEQUENCES AND SERIES: COMPOUND INTEREST AND ANNUITIES

1 The Power of Compounding

Intra-Industry Trade in Intermediate Products, Pollution and Internationally Increasing Returns *

Standard Deviations for Normal Sampling Distributions are: For proportions For means _

B = A x z

CreditRisk + Download document from CSFB web site:

Solutions to Problem Sheet 1

The material in this chapter is motivated by Experiment 9.

Subject CT5 Contingencies Core Technical. Syllabus. for the 2011 Examinations. The Faculty of Actuaries and Institute of Actuaries.

Introduction to Probability and Statistics Chapter 7

of Asset Pricing APPENDIX 1 TO CHAPTER EXPECTED RETURN APPLICATION Expected Return

The Time Value of Money in Financial Management

Subject CT1 Financial Mathematics Core Technical Syllabus

Non-Inferiority Logrank Tests

x satisfying all regularity conditions. Then

DESCRIPTION OF MATHEMATICAL MODELS USED IN RATING ACTIVITIES

Confidence Intervals. CI for a population mean (σ is known and n > 30 or the variable is normally distributed in the.

0.07. i PV Qa Q Q i n. Chapter 3, Section 2

Correlation possibly the most important and least understood topic in finance

Unbiased estimators Estimators

An Empirical Study of the Behaviour of the Sample Kurtosis in Samples from Symmetric Stable Distributions

2. The Time Value of Money

Analysis III (BAUG) Assignment 1 Prof. Dr. Alessandro Sisto Due 29 of September 2017

Section 3.3 Exercises Part A Simplify the following. 1. (3m 2 ) 5 2. x 7 x 11

Systems Analysis Laboratory Research Reports E16, June 2005 PROJECT VALUATION IN MIXED ASSET PORTFOLIO SELECTION

Introduction to Financial Derivatives

CHAPTER 8: CONFIDENCE INTERVAL ESTIMATES for Means and Proportions

ad covexity Defie Macaulay duratio D Mod = r 1 = ( CF i i k (1 + r k) i ) (1.) (1 + r k) C = ( r ) = 1 ( CF i i(i + 1) (1 + r k) i+ k ) ( ( i k ) CF i


KEY INFORMATION DOCUMENT CFD s Generic

CAUCHY'S FORMULA AND EIGENVAULES (PRINCIPAL STRESSES) IN 3-D

1 + r. k=1. (1 + r) k = A r 1

CAPITAL PROJECT SCREENING AND SELECTION

Methodology on setting the booking prices Project Development and expansion of Bulgartransgaz EAD gas transmission system

Hopscotch and Explicit difference method for solving Black-Scholes PDE

STAT 135 Solutions to Homework 3: 30 points

Transcription:

OPERATIONS RESEARCH doi 0.87/opre.090.079ec e-copaio ONLY AVAILABLE IN ELECTRONIC FORM ifors 009 INFORMS Electroic Copaio Equilibriu Capacity Expasio Uder Stochastic Dead Growth by Alfredo Garcia ad Zhijiag She, Operatios Research, doi 0.87/opre.090.079.

ec e-copaio to Author: Alfredo Garcia ad Zhijiag She Proofs of Stateets EC.. Proof of Theore. (cotiued) EC... Regio S I the paper, we have checked that there is o (strictly) profitable deviatio fro y whe κ > ρ(+g) θg i regio S. The proof follows by checkig this coditio i all reaiig regios. EC... Regio S If the iitial relative capacities (k, 0 k) 0 are i this regio, there will be o ivestet accordig to strategy y. If fir deviates by choosig a target of capacity k > δk, 0 the the derivative of fir s objective fuctio is: D(k ) = κ + βθ(), δk 0 ) + β( θ) (k, δk 0 ) (EC.) If (k 0, k 0 ) {k 0 k δ < k0 i (+g), i =, }, the assuptio δ < guaratees the relative capacities 3δ +g after depreciatio will be regio S. They are either (k, δk 0 ) if o growth o the dead side or, δk0 ) if dead grows. The forer capacity pair is i regio IV ad the latter oe is i regio I. Thus we have: (k, δk k ) 0 = δκ, δk 0 ) = R, δk 0 ) + δ κ Substitutig i EC. we obtai the followig expressio for the derivative: D(k ) = ( βδ)κ + βθ() R, δk 0 ) Sice R k < 0, R < 0 i regio I ad δk 0 k > k, k > δk 0 > k,so it is true that Thus, for all (k 0, k 0 ) {k 0 k δ D(k ) < ( βδ)κ + βθ() R ( k, k ) = 0 < k0 i (+g), i =, } fir ca ot be better off by ivestig. 3δ

e-copaio to Author: Alfredo Garcia ad Zhijiag She ec3 If (k, 0 k) 0 {k 0 ki 0 > (+g), i =, }, the relative capacities after depreciatio will reai i 3δ regio IV o atter whether dead growth or ot. We kow the short ter equilibriu reveue R is a costat i regio IV, so fir has o icetive to ivest i this case. Therefore, it is ot profitable for fir to ivest i regio S. EC..3. Regio S 3 If the iitial relative capacities (k, 0 k) 0 are i this regio, strategy y prescribes o ivestet for fir. We defie fir s ivestet target as ϕ(k) 0 which is solely decided by the relative capacity level of fir. If fir ivests to reach a relative capacity level k, we have: The derivative of the objective is: (k, δk k ) 0 = R (k, δk k ) 0 + δκ, δk 0 ) = R, δk 0 ) + δ κ D(k ) = ( βδ)κ + βθ() R, δk 0 ) + β( θ) R (k, δk 0 ) (EC.) We ow show that it is ever optial for fir to choose a capacity level i regio I uder our assuptio > ρ(+g). The derivative for this case is: κ θg D(k ) = ( βδ)κ + β βδk0 β ( θg) ( θg)k which is decreasig i k. Whe k reaches the right boudary of regio I, that is k = δk 0, D(k ) is iiized: ( θg) D(k ) > ( βδ)κ β + Sice δk 0 > k for ay iitial k 0 capacity i regio S 3, we have: ( θg) D(k ) > ( βδ)κ β + 3( θg) βδk 0 So it is ever optial for fir to choose a capacity level i regio I. 3( θg) βk > 0 Based o the relative capacities (k, δk) 0 ad, δk0 ), EC. ay have differet expressios for a give k 0.

ec4 e-copaio to Author: Alfredo Garcia ad Zhijiag She Case : If (k, δk 0 ) IV ad +g, δk0 +g ) I: D (k ) = ( βδ)κ + βθ[ (k + δk 0 )] Case : If (k, δk 0 ) II ad +g, δk0 +g ) I: D (k ) = ( βδ)κ + βθ[ (k + δk 0 )] + β( θ) ( k ) Notice that D (k ) > D (k ). That eas for the sae δk 0, if the locally optial poit appears i Regio IV (D (k ) = 0) the D (k ) ust be positive. It is optial for fir to ivest to ake the capacity pair (k, δk 0 ) eter regio IV. Case 3: If (k, δk) 0 II ad, δk0 ) II:, the derivative of the objective is: D 3 (k ) = ( βδ)κ + θ β β( + θ)k Case 4: Ad if (k, δk 0 ) IV ad +g, δk0 +g ) II: D 4 (k ) = θ β ( βδ)κ β( + θ)k Sice there will be o ivestet i regio S, fir s ivestet level is bouded by (+g) 3. For a fixed δk 0, if there is oly oe locally optial k (that is D(k ) = 0), this k ust be the optial target of capacity ivestet for fir ; If for exaple, D 3 (k ) > 0 ad D 4 (k ) < 0, the optial level of ivestet for fir will be lie k = ; If there are ore tha oe locally optial level 3 of k, we ust copare the value of objectives correspodig to each targets levels i order to idetify the real optial level of ivestet level of capacity for fir. EC..4. Regio S 4 If the iitial relative capacities (k, 0 k) 0 are i this regio, fir will ivest to ake its relative capacity i the ext period equal to ϕ(k). 0 We are goig to check if it is profitable for fir to deviate by akig a little ivestet i this regio. Fir will ivest oly if the relative capacities after ivestet, ϕ(k0 ) ) are i regio I, otherwise fir s oe period reveue is oly decided

e-copaio to Author: Alfredo Garcia ad Zhijiag She ec5 by fir s relative capacity. (accordig to the clai i the previous part, (k, ϕ(k 0 )) ca ot be i regio I). We have: The derivative of the objective is: (k, ϕ(k k )) 0 = δκ, ϕ(k0 ) ) = R, ϕ(k0 ) ) + δ κ D(k ) = ( βδ)κ + βθ() R, ϕ(k0 ) ) + β( θ) R (k, ϕ(k 0 )) (EC.3) Case : If the relative capacities (k, ϕ(k 0 )) IV ad, ϕ(k0 ) ) I: we kow that for fir Thus, D (ϕ(k 0 )) = ( βδ)κ + βθ[ (δk0 + ϕ(k 0 ))] = 0 ϕ(k 0 ) = () [ ( βδ)κ ] δk0 βθ ad if fir ivests to reach k > δk 0, the D(k ) = ( βδ)κ + βθ[ (k + ϕ(k 0 ))] = [ ( βδ)κ + βθ( 4k δk 0 )] < [ ( βδ)κ + βθ( 3k )] < 3k [ ( βδ)κ + βθ( )] = 0 Hece, it is ot profitable for fir to ivest i this case. Case : If (k, δk) 0 III ad, δk0 ) I: we kow fro the previous part that the ivest- et target for fir is: D (ϕ(k 0 )) = ( βδ)κ + βθ[ (ϕ(k0 ) + δk 0 )] + β( θ) ( ϕ(k0 )) = 0 Thus, ϕ(k 0 ) = ( + ( θ)(+g) θ ) ( [ ) β( + θ) ( βδ)κ]() δk 0 βθ

ec6 e-copaio to Author: Alfredo Garcia ad Zhijiag She ad k + ϕ(k 0 ) > δk 0 + ϕ(k 0 ) = [ β( + θ) ( βδ)κ]() [ + βθ > [ β( + θ) ( βδ)κ]() βθ > 3k ( θ)() ]ϕ(k 0 θ ) Hece, D(k ) = ( βδ)κ + βθ[ (k + ϕ(k 0 ))] < 3k [ ( βδ)κ + βθ( )] = 0 ad it is ot profitable for fir to ivest i this case. EC.. Proof of Theore, whe ρ < κ ρ(+g) θg Siilar with the proof i the previous part, based o the iitial relative capacity prior depreciatio, we partitio the etire relative capacity space R + ito 4 regios slightly differet with before as figure EC. idicates: k S 3 S ( + g) 3 S S 4 * k 3δ ( + g) 3δ k Figure EC. The partitio of relative capacity space whe ρ < κ ρ(+g) θg.

e-copaio to Author: Alfredo Garcia ad Zhijiag She ec7 S = {k 0 ki 0, i =, } 3δ S = {k 0 ki 0 > (+g), i =, } 3δ S 3 = {k 0 k 0 < k}\(s 0 S ) S 4 = {k 0 k 0 > k}\(s 0 S ) We propose a ivestet strategy cobiatio ŷ whe ρ < ρ(+g), forally described as κ θg follows where ax{k δk ; 0} k i S ŷ (k) = ax{ϕ(k) δk ; 0} k i S 3 0 otherwise k = [β ( βδ)κ]() 3β( θg) ad ϕ(k) is a fuctio to be defied later. Because of syetry, it is eough to check the optiality of strategy ŷ whe fir follows strategy ŷ. We defie BR (k ) = k, BR (k ) = (+g) k whe k k ad defie BR (k ), BR (k ) accordigly whe k > k. EC... Regio S Suppose the iitial relative capacities are such that δk 0 i k ad fir s capacity at the begiig of the ext period after ivestet is k = δk 0 + y. Accordig to strategy ŷ, fir s decisio is to achieve a relative capacity level of k. So fir s objective fuctio is ax{r k (k, 0 k) 0 (k δk)κ 0 + β[θ()v, k ) + ( θ)v (k, k )]} The derivative D(k ) of the objective fuctio above with respect to k is: D(k ) = κ + βθ(), k ) + β( θ) (k, k ) (EC.4) If fir chooses a target capacity ivestet level k k, v (k, k ) = R (k, k ) (k δk )κ + C(k ) v, k ) = R, k ) (k δk )κ + C(k ) where C(k ) = β[θ()v ( k +g, k +g ) + ( θ)v (k, k )] is a costat oly related to the capacity level i equilibriu k. Thus, (k, k ) = R (k, k ) + δκ

ec8 e-copaio to Author: Alfredo Garcia ad Zhijiag She, k ) = R, k ) + δ κ Substitutig i EC.4 we obtai the followig expressio for the derivative: Note that D(k ) = ( βδ)κ + βθ() R, k ) + β( θ) R (k, k ) = ( βδ)κ + βθ[ ( k + k )] + β( θ)[ (k + k )] D(k ) = ( βδ)κ + βθ() R, k ) + β( θ) R (k, k ) k =k k = ( βδ)κ + βθ( 3k ) + β( θ)( 3k ) = β ( βδ)κ 3βk ( θg ) = 0 k [β ( βδ)κ]() = 3β( θg) k is i regio I oly if ρ < ρ(+g), where ρ = βδ. κ θg β The differece betwee D(k ) ad D(k ) is D(k ) D(k ) = βσ( θg )(k k ) We have that D(k ) > D(k ) = 0 whe k < k ad D(k ) < D(k ) = 0 for all k < k k δ. If the iitial relative capacities satisfy k 0 i S {k 0 i k < δk 0 k () ad k 0 k 0 }, ad fir chooses a target capacity ivestet level k, fir will ot ivest accordig to strategy ŷ. We have k =k v (k, δk 0 ) = R (k, δk 0 ) (k δk )κ + C(k ) v, δk 0 ) = R, δk 0 ) (k δk )κ + C(k ) Siilar to EC.4, the first order coditio of fir s objective is: D(k ) = κ + βθ()[ R, δk 0 ) + δκ ] + β( θ)[ R (k, δk 0 ) + δκ] = ( βδ)κ + βθ() R, δk 0 ) + β( θ) R (k, δk 0 )

e-copaio to Author: Alfredo Garcia ad Zhijiag She ec9 = ( βδ)κ + βθ[ ( k + δk0 )] + β( θ)[ (k + δk 0 )] = 0 We get the ivestet target for fir i this case k (k 0 ) = [β ( βδ)κ]() β( θg) δk0 = 3k δk0 If the iitial relative capacities satisfy k 0 i S {k 0 i k < δk 0 k () ad k 0 < k 0 }, ad fir chooses a target capacity ivestet level k slightly greater tha δk 0, we have v (k, k (k 0 )) = R (k, k (k 0 )) (k δk )κ + C(k ) v, k (k 0 ) ) = R, k (k 0 ) ) (k δk )κ + C(k ) The derivative of fir s objective is: D(k ) = κ + βθ()[ R, k (k 0 ) ) + δκ ] + β( θ)[ R (k, k (k 0 )) + δκ] Sice k (k 0 ) = 3k = ( βδ)κ + βθ[ σ( k + k (k 0 ) )] + β( θ)[ σ(k + k (k 0 ))] δk0, the derivative becoes D(k ) = ( βδ)κ + βθ[ σ( k + k (k 0 ) )] + β( θ)[ σ(k + k (k 0 ))] = β ( βδ)κ βσ( θg )(k + 3k δk0 ) < β ( βδ)κ βσ( θg )( 3δk0 + 3k ) < β ( βδ)κ 3βσk ( θg ) = 0 So it is ot profitable for fir to ivest i this case. The siilar aalysis ca be carried out for iitial relative capacities i regio k 0 i S {k 0 i k (+ g) t < δk 0 k () t+ ad k 0 k 0 } ad k 0 i S {k 0 i k () t < δk 0 k () t+ ad k 0 < k 0 } where t =,, 3,... Therefore, we coclude that for all the iitial relative capacities (k 0, k 0 ) i regio S, k (k 0 ) ad k (k 0 ) describes the optial ivestet target for each fir. If both k 0 ad k 0 are less tha k /δ,

ec0 e-copaio to Author: Alfredo Garcia ad Zhijiag She they both ivest to reach k at the begiig of ext period; if both k 0 ad k 0 are greater tha the correspodig level described by k (k) 0 ad k (k), 0 there will be o ivestet at all; for the other cases, oly the fir with less relative capacity will ivest to a level of capacity decided by the iitial capacity of the oppoet fir. EC... Regio S If (k, 0 k) 0 {k 0 ki 0 > (+g), i =, }, the relative capacities after depreciatio will reai at regio 3δ IV o atter whether dead growth or ot. We kow the short ter equilibriu reveue R is a costat i regio IV, so o fir has icetive to ivest for this case. EC..3. Regio S 3 Sice we already kow the optial ivestet strategy ŷ i regio S, we ca idetifiy the states trajectories startig fro ay iitial relative capacities i regio S. Therefore, the expected value v (k) i regio S is kow. We divide regio S 3 ito a series of horizotal stripes I t = {k 0 i k 0 i S 3 ad 3 ()t < δk 0 ( + 3 g)t+ } where t = 0,,,... If (k, 0 k) 0 I 0, we defie fir s ivestet target as ϕ(k) 0 which is solely decided by the iitial relative capacity level of fir. If fir ivests to reach a relative capacity level k, the derivative of the objective is: D(k ) = κ + βθ(), δk 0 ) + β( θ) v (k, δk 0 ) (EC.5) Sice there will be o ivestet i regio S, fir s ivestet level is bouded by (+g) 3. For a fixed δk 0, if there is oly oe locally optial k (that is D(k ) = 0), this k ust be the optial target of capacity ivestet for fir ; If for exaple, D(k ) > 0 whe k < 3σ ad D(k ) < 0 whe k >, the optial level of ivestet for fir will be lie k 3σ = ; If there are ore tha 3σ oe locally optial level of k, we ust copare the value of objectives correspodig to each targets levels i order to idetify the real optial level of ivestet level of capacity for fir. After we get fir s ivestet target as ϕ(k 0 ) i regio I 0, we are able to calculate the expected value fuctio of fir v (k) i this stripe. The we ca idetify fir s ivestet target whe

e-copaio to Author: Alfredo Garcia ad Zhijiag She ec (k, 0 k) 0 I. Fir s optial ivestet target ϕ(k) 0 for those stripes with low idex ubers I 0, I,... ay ot be cotiuous alog k, 0 the reaso for that is because the short ter reveue is oly piecewise cotious i the capacity space. Let s exaie ϕ(k) 0 for I t with relatively large idex uber. If fir ivests to reach a relative capacity level k, we have: The derivative of the objective is: (k, δk k ) 0 = R (k, δk k ) 0 + δκ, δk 0 ) = R, δk 0 ) + δ κ D(k ) = ( βδ)κ + βθ() R, δk 0 ) + β( θ) R (k, δk 0 ) Case : If (k, δk) 0 II ad, δk0 ) II:, the derivative of the objective is: Case : If (k, δk 0 ) IV ad +g, δk0 +g ) II: D (k ) = θ β ( βδ)κ β( + θ)k D (k ) = θ β ( βδ)κ β( )k Followig the guidlies we just etioed above, if there is oly oe local axiu, this is the optial target ivestet level; if D (k ) > 0 whe k < 3 ad D (k ) > 0 whe k > 3, ϕ(k0 ) = (+g) ; if D 3 (k ) > 0 whe k < ad D 3 (k ) < 0 whe k >, 3 ϕ(k0 ) = ; ad if there are two 3 local axia levels, we eed to copare the expected value fuctio explicitly. EC..4. Regio S 4 If the iitial relative capacities (k, 0 k) 0 are i this regio, fir will ivest to ake its relative capacity i the ext period equal to ϕ(k). 0 We are goig to check if it is profitable for fir to deviate by akig a little ivestet i this regio. Fir will ivest oly if the relative capacities after depreciatio ad after dead growth ( δk0, ϕ(k0 ) ) are i regio I, otherwise fir s oe period

ec e-copaio to Author: Alfredo Garcia ad Zhijiag She reveue is oly decided by fir s relative capacity. Siilar with the aalysis for regio S 3, we divide regio S 4 ito a series of vertical stripes H t = {k 0 i k 0 i S 4 ad 3 (+g)t < δk 0 3 (+g)t+ } where t = 0,,,... For the first stripe H 0, if (k 0, k 0 ) H 0 ad fir chooses a target level of ivestet k 3δ we have: v (k, ϕ(k 0 )) = R (k, ϕ(k 0 )) + β[θ()v, ϕ(k0 ) ) + ( θ)v (k, ϕ(k 0 ))] Thus, (k, ϕ(k 0 k )) = β( θ) [ R (k, ϕ(k k )) 0 + βθ(), ϕ(k0 ) )] The derivative of the objective is: D(k ) = κ + β( θ) [β( θ) R (k, ϕ(k k )) 0 + βθ(), ϕ(k0 ) )] (EC.6) If the relative capacities +g, ϕ(k0 ) +g ) are above the coectio of k (k 0 ) ad k (k 0 ), fir will ot ivest accordig to the optial strategy ŷ i regio S. So we have The derivative i EC.6 becoes, ϕ(k0 ) ) < κ D(k ) < κ + β( θ) [β( θ) R (k, ϕ(k 0 )) + βθκ] (i) If the relative capacities +g, ϕ(k0 ) +g ) are uder the coectio of k (k 0 ) ad k (k 0 ), fir will ivest to k (k 0 ) or k accordig to the optial strategy ŷ i regio S. So we have The derivative i EC.6 becoes, ϕ(k0 ) ) = R, ϕ(k0 ) ) + δ κ D(k ) = ( βδ)κ + βθ() R, ϕ(k0 ) ) + β( θ) R (k, ϕ(k 0 )) (ii) Case : If the relative capacities (k, ϕ(k 0 )) IV ad D(k ) < κ +, ϕ(k0 ) βθκ = ( β) < 0 β( θ) ) I: the derivative i (i) is

e-copaio to Author: Alfredo Garcia ad Zhijiag She ec3 Sice it is true for fir that ( βδ)κ + βθ[ (δk0 + ϕ(k 0 ))] = 0 Thus, ϕ(k 0 ) = () [ ( βδ)κ ] δk0 βθ ad the derivative i (ii) becoes D(k ) = ( βδ)κ + βθ[ (k + ϕ(k 0 ))] < [ ( βδ)κ + βθ( 3k )] < 3k [ ( βδ)κ + βθ( )] = 0 Hece, it is ot profitable for fir to ivest i this case. Case : If (k, δk) 0 III ad, δk0 ) I: the derivative i (i) is Sice it is ture for fir that: D(k ) < κ + βθκ = ( β) < 0 β( θ) ( βδ)κ + βθ[ (ϕ(k0 ) + δk 0 )] + β( θ) ( ϕ(k0 )) = 0 Thus, ad ϕ(k 0 ) = ( + ( θ)(+g) θ ) ( [ ) β( + θ) ( βδ)κ]() δk 0 βθ k + ϕ(k 0 ) > δk 0 + ϕ(k 0 ) = [ β( + θ) ( βδ)κ]() [ + βθ > [ β( + θ) ( βδ)κ]() βθ > 3k ( θ)() ]ϕ(k 0 θ ) Hece, the derivative i (ii) is D(k ) = ( βδ)κ + βθ[ (k + ϕ(k 0 ))]

ec4 e-copaio to Author: Alfredo Garcia ad Zhijiag She < 3k [ ( βδ)κ + βθ( )] = 0 ad it is ot profitable for fir to ivest i this case. S. Case 3: If (k, δk) 0 I ad, δk0 ) I: the aalysis is the sae as what we did i regio The siilar arguet holds for (k 0, k 0 ) H t, t =,,... I Suary, for ay iitial relative capacities (k 0, k 0 ), there exists a MPE strategy ŷ whe ρ < κ ρ(+g) θg so that ay deviatio fro which is ot profitable. EC.3. Uiqueess of y Cosider the class of ivestet strategies, say ỹ(k), that prescribe ivestets ỹ i (k) = k i δk i wheever δk i k i ad k k. Suppose the iitial capacity stock k 0 is such that δki 0 k i. Fir i s objective fuctio is ax{r i (k 0 ) (k i δk 0 k i i )κ + β[θ()v i ( k i, kj ỹ) + ( θ)v i(k i, k j ỹ)]} The (ecessary) first order coditio for equilibriu for i {, } is: ( βδ)κ = βθ() R i k i ( k i, kj ) + β( θ) R i ( k i, k k j ) i Without loss of geerality, it is eough to cosider the coditio where k i k j. The argial profits are as follows: Regio I: { R i k i ( k i, k j ) = ( k i + k j ) R i k j ( k i, k j ) = ( k i + k j ) Regio II: { R i k i ( k i, k j ) = k i R i k j ( k i, k j ) = 0 Regio IV : { R i k i ( k i, k j ) = 0 R i k j ( k i, k j ) = 0

e-copaio to Author: Alfredo Garcia ad Zhijiag She ec5 Note that best reply fuctios for fir i ad j are liear with differet slopes whe ( k i, k j ) ad ( k i +g, k j +g ) {I, IV }. So there ca oly exists a syetric equilibriu i these cases. If ( k i, k j ) II ad ( k i, k j ) I. The best reply of fir i is: k (i) j The best reply of fir j is: = βθ [ θ β( + θ) ( βδ)κ β( + θ) k i ] k (j) j The slope of fir i s best reply equals = ( ρκ θ ) k i ( θ ()( θ) + θ) = θ θ which is steeper tha fir j s best reply. We check for a itersectio of best reply aps whe ( k i, k j ) II ad ( k i +g, k j +g ) I. Whe k i = 3, k i locates at the right boudary of this regio, the correspodig k j fro fir i s best reply is k (i) j The correspodig k j fro fir j s best reply is Hece, the differece is: = ()ρκ ( + θ + 5θg) 6θ θ k (j) j = 6 ( + 3g) ()ρκ θ k (i) (j) j k j = ()ρκ ( θ + θg) 6θ θ Reeber that we require κ > ρ(+g) θg i Theore, we get κ < θg. Plug this relatio ito the ρ(+g) above expressio, this leads to k (i) (j) j k j > θg ( θ + θg) 6θ θ = ()( θ) 0 6θ Therefore, fir i s best reply is always above fir j s best reply, so these two best replies will ot itersect i this regio. If (( k i, k j ) II ad ( k i +g, k j +g ) II)or (( k i, k j ) IV ad ( k i +g, k j +g ) II), we have that the right side of first order coditio for equilibriu for fir j is always equal to 0. This eas fir j has o icetive to ivest to these regios.

ec6 e-copaio to Author: Alfredo Garcia ad Zhijiag She EC.4. Algebraic Derivatio i 5. H() + F (r θg) = θ()r ( k (), k EC.5. Etry Deterrece = θ[ k () ]k () + ( θ) ) + ( θ)r (k (), k ) ρκk () ( ) ρκk () + = θ[ + ( ρκ θ ) ρκ θ )]k () + ( θ) = θ ( ρκ ) ( ) k () + ( θ) + θ + ( ρκ ) ( ) θ = θ() + ( θ) + + ( ) + The paraeters i this uerical illustratio are κ =, β = 0.87, δ =, θ = 0.6, g = 0., F = ad let free. With this choice of paraeters, > 5, provided >.5. Assue icubet firs wat to deter the etry by aitaiig a level of capacity k > k (). Note that i order to deter etry, k does ot eed to be greater tha (+g). Figure EC. shows the state trajectory whe etry occurs. Etry is ot profitable if ad oly if: ax x { F κx + β[θ()v( x, ( ) () Etry occurs whe there is dead growth. Sice ( ) (+g) ( ) ) + ( θ)v(x, )]} 0 (EC.7) ( )k (), the firs will retur to syetric capacity levels equal to k () just oe period after the dead growth has bee ( ) k' * ( ) k ( ) ( ) K i * * ( k ( ), ( ) k ( )) x + K i Figure EC. The Capacity Trajectory Whe Etry Occurs.

e-copaio to Author: Alfredo Garcia ad Zhijiag She ec7 observed. Sice there are firs havig a capacity stock greater tha k (), the optial scale x for the etrat (i.e. the -th fir) ust be saller tha k (). We have v(x, ad ( ) ) = R(x, ( ) x ) + β[θ()v(, ( ) () ( ) ) + ( θ)v(x, )] (EC.8) x ( ) v(, () ) = R( x ( ), () ) (k () x )κ (EC.9) +β[θ()v( k (), ( )k () ) + ( θ)v(k (), ( )k ())] Fro EC.7-EC.9, we copute the derivative of the left side of EC.7: ( ) R(x, ) β( θ) x + βθ() R( x, ( ) ) +g (+g) ( β)κ 0 x (EC.0). Suppose (x, ( ) ) falls i regio II. I this case, the derivative becoes β( θ)( ( ) x+ x) + βθ[ ] ( β)κ +g = 0.3 + 0.436.009x This derivative equals to 0 whe x = 0.44 0.88. So the optial level is x = i{0.44 0.88, 6 }. Suppose (x, ( ) ) falls i regio IV. I this case, the derivative becoes ( ) x+ βθ[ ] ( β)κ +g = 0.74 0.87x 0.3 This derivative is ootoe decreasig i x ad equals to 0 whe x = 0.494 + 0.. So the optial level of x is: x = ax{ 0.494 + 0., 6 } I suary, the optial level of x should be 0.44 0.88.4943 <.735 x =.735 < 4.480 6 0.494 + 0. 4.480 First, let s see if the etrat ca acquire positive profit whe x = 0.44 0.88. The expected value of the etrat is: F κx + β[θ()v( x F κx + β( θ) β( θ), ( ) +g (+g) = R(x, ( ) ) + ( θ)v(x, ( ) )] ) + βθ(+g) β( θ) v( x +g, ( ) (+g) )

ec8 e-copaio to Author: Alfredo Garcia ad Zhijiag She Also, x ( ) v(, () ) = R( x ( ), () ) (k () x )κ + β β( + θg) [θ()r i(k ()) + ( θ)r i ( k () ( θ)g ) κk ()] =.0839( 0.935)( + 0.304) Therefore, the expected value of the etrat whe.4943 <.735 is: F κx + β( θ) β( θ) which is positive whe >.5096. R(x, ( ) ) + βθ(+g) (.0839( 0.935)( + 0.304)) β( θ) =.0609(.5096)( +.35) I a aalogous way, we ca fid out the rage of which ight deter etry for the other two cases: whe x =,.the expected value of the etrat is:.0566(.5097)( +.64) which is 6 always positive sice.735 < 4.480. whe x = 0.494+0., the expected value of the etrat is:.0596(.55)(+.37) which is positive sice > 4.480 i this case. This result idicates that whe >.5096, it is always profitable for a ew fir to eter the arket. The icubets ca ever deter the occurrece of etry by aitaiig a higher level of capacity. Ackowledgets The authors gratefully ackowledge all the coets ad observatios received fro the referees which have iproved the paper sigificatly. Refereces See refereces list i the ai paper.