lec3 dynamic duopoly continued - University of...

Post on 24-Sep-2020

2 views 0 download

Transcript of lec3 dynamic duopoly continued - University of...

Lecture 3: Dynamic Cournot Duopoly Continued

Technology

• Ki capital of firm i

• Qi output of firm i

• qi =KiQioutput per unit of capital

• c(q) cost per unit of capital when output intensity is q. c0 > 0,

c00 > 0.

• Transition Ki,t+1 = Qi,tσ (σ = 1 − δ) so Leontief in output

and next period capital.

Markov Perfect Equilibrium in this Model

• Payoff Relevant State: (K1,K2)

• A MPE is set of policy functions qi(K1,K2), i = 1, 2, andvalue functions wi(K1,K2), i = 1, 2, such that

w1(K1,K2) = maxq1≥0

P (q1K1 + q2(K1,K2))q1K1 − c(q1)K1

+βw1(σq1K1, σq2(K1,K2)K2)

and q1(K1,K2) solves the above program and the analogousconditions hold for for the policy function and value functionof firm 2.

• Firm 1’s current choice of q1 on firm 2’s next period q2in thecurrent period on firm 2’s choice of q02 next period?

q02 = q2(σq1K1, σq2(K1,K2)K2)

In firm 1’s decision making this is buried in βw2(K01,K

02)

• Tricky problem, nibble pieces off, one step at time.

Start with Analysis of β = 0 case.

(Still get intesting dynamics, get lot’s of intuition that does

generalize)

• Given (K1,K2), solve the (asymmetric) Cournot duopoly prob-lem

• Claim: if K1 > K2 then q1 < q2, but q1K1 > q2K2.

–FONC for two firms

P + P 0q1K1 − c0(q1) = 0

P + P 0q2K2 − c0(q2) = 0

Suppose instead that q1 ≥ q2.

⇒ c0(q1) ≥ c0(q2)

⇒ P 0q1K1 ≥ P 0q2K2

⇒ K1 ≤ K2, a contradiction.

• Claim market shares converge to equality.

•K01

K02

=q1K1(1− δ)

q2K2(1− δ)

=q1K1q2K2

<K1K2

But

1 <K01

K02

• So converge to 50-50 monotonically.

–Kydland, Dominant firm literature

• Intuition?

• Suppose β > 0

–analytic results difficult

–will go to computer and work this out

–Suppose commit to sequence of outputs. Does this matter?

Look at T = 2 case.

Comment About the Role of Commitment

• MPE equilibrium very differernt from outcome of simultaneousmove game where firm one and two pick vectors (q11, q12, q13, ...)

and (q21, q22, q23, ...)

Benchmark Case of Perfect Competition Steady State

• Suppose agents take as given a constant price p. (approxi-

mately the same as if there are a large number of firms).

• Let v be the discounted value of owning one unit of capital atthe beginning of a period

v = maxq

pq − c(q) + βσqv

where

σ = 1− δ

• FONC

p− c0(q) + βσv = 0 (1)

• In a stationary equilibrium,

σq = 1

q∗ =1

σ

• v∗ solves

v∗ = pq∗ − c(q∗) + βσq∗v∗

= pq∗ − c(q∗) + βv∗

so

v∗ =pq∗ − c(q∗)1− β

• From the FONC

p = c0(q∗)− βσv∗

• Plugging in the formula for v∗ yields

p = c0(q∗)− βσpq∗ − c(q∗)1− β

Solving for p yields the stationary competitive price

p∗C = (1− β)c0(q∗) + βσc(q∗).

• Q∗C be the stationary competitive output

• x∗C = σQ∗C be the stationary competitive capital level.

Pure Monopoly.

• The state variable is K at the beginning of period capital.

Let w(K) be discounted maximized monopoly profit. This

solves

w(K) = maxq

P (Kq)Kq −Kc(q) + βw (σKq)

• The FONC is

PK + P 0K2q −Kc0 + βσKdw

dK= 0

• Dividing by x,

P + P 0Kq − c0 + βσdw

dK= 0

• Use the envelope theorem to verify that

dw

dK= qc0(q)− c(q)

(Think of Q as the choice variable....).

• Plugging this into the first-order condition and evaluating atthe steady state output level q∗ = 1

σ yields

p+ P 0qK − c0 + βσhqc0 − c

i= 0

or

p+ P 0q∗K = (1− β) c0 + βσc

= P ∗C.

• Let K solving the above be denoted K∗M . .

• Now calculate the equilbrium off the steady state

A Technical Aside

Numerical Solutions of Dynamic Programming Problems

Monopoly Problem

• Statement of problem. w(K) value function and q(K) is

policy function. Contraction mapping: Let w0 be value

function beginning next period. Then

w1(K) = maxqP (Kq)Kq −Kc(q) + βw0 (σKq) .

A solution is where w1(K) = w0(K) for all K.

• Iterate

• How do numerically? Need an approximation for w0.

• Discretize? Works well with single agent decision theory. For

duopoly problem though continuity is useful.

• Polynomial approximation.

Example with Linear Approximation

1. Start with approximation

w0(K) = α0 + β0K

2. Take a set of m evaluation points K = {K1, K2, ..., Km}

3. Solve problem at each of this points with w0(K) instead of

w0(K).

w1,i = maxqP³Kiq

´Kiq − Kic(q) + βw0

³σKiq

´.

4. Yields a vector W1 = (w1,1, w1,2, ....w1,m)

5. Use OLS to determine a new approximationÃα1β1

!=

³X 0X

´−1X 0W1

X = 1˜K

6. Iterate until obtain convergence in (αt, βt)

General Polynomial Approximation

• Chebyshev polynomials (in class of orthogonal polynomials)

• Defined on range x ∈ [−1, 1]

Tn(x) = cos(n cos−1 x)

-1.5

-1

-0.5

0

0.5

1

1.5

-1 -0.5 0 0.5 1

T0T1T2T3T4

Figure 1:

Recipe in Judd

• Step 1: Evaluation points

zk = − cos(2k − 12m

π), k = 1, ...,m

• Step 2: Adjust the notes to the [a,b] interval (here a =

.5K∗M, b = 1.5K∗M)

xk = (zk + 1)µb− a

2

¶+ a, k = 1, ...,m

• Step 3: Evaluate w(x) at the approximation nodes

wk = w(xk), k = 1, ...,m

• Step 4: Compute the Chebyshev coefficients (remember Tiorthogonal)

ai =

Pmk=1 wkTi(zk)Pmk=1 Ti(zk)

2

• To arrive at the approximation

w(x) =nXi=0

aiTi(2x− a

b− a− 1)

Hints for Duopoly Problem

• (a0, ...an) coefficient vector for the value function v1(K1,K2)approximation

• (b0, ...., bn) coefficient vector for the policy function q1(K1,K2)approximation.

• Use Judd’s techniques for approximation in R2 (page 238)

• You need to iterate on q1 as well as v1 since firm 1 takes firm

2’s action as given in the problem (and q2(x, y) = q1(y, x)).

Going to do some work. Think about the lesson you are supposedto be learning.

1. Industry with adjustment costs and long-run constant returnsto scale, large firms decline. So initial asymmetry gets smoothedout in the long run.

2. Logic very clear. The smaller a firm’s market share, the lessthey take into account the negative impact of expansion ofoutput on price.

3. So if have asymmetry, need something else. (Like scale economies,heterogeneity in productivity)

4. What happens with mergers? See Gowrisankaran and Holmes(2004) Rand paper.