Introduction to Endogenous Growth Models 55

Introduction to Endogenous Growth Models

 Paul Romer’s 1986 model and Robert Lucas’s (1988) human capital model.
 These models get around the diminishing marginal returns to “capital” assumption by
broadening the definition of capital to include knowledge or human capital, both of
which may have positive externalities.
 We need some kind of external effects in order to have a model in which
o individual firms do not have increasing overall returns to scale, so they do not
expand infinitely and become economy-wide monopolies
o the economy as a whole has increasing returns to scale, so that returns to
“capital” or “produced inputs” can be constant
 Endogenous growth model have several features that economists have found attractive
o They endogenize key parameters of the model such as g
o They can explain lack of convergence
o They allow s and related policy variables to affect the growth rate of GDP, not
just the level of the growth path
 Text book begins with a simplified model of knowledge production via research and
development in Chapter 3.
o Uses constant saving assumption as in Solow model
o Incorporating Ramsey saving model does not change basic dynamics
 Key characteristic leading to endogenous growth: constant returns to scale in produced
inputs.
o In Solow and Ramsey models, capital was only produced input and had
diminishing returns

David Romer’s R&D model

Dynamics and behavioral assumptions
 Economy has two sectors: goods-producing sector and R&D (knowledge-producing)
sector
 Each sector uses labor and capital
o aL and aK are the shares of labor and capital allocated to the knowledge sector
o These should be determined by choices of owners of labor and capital allocating
them to their highest return
o Romer simplifies the model by taking these to be exogenous
o Econ 454 studies models in which the rewards to capital and labor in the two
sectors are explicitly modeled and these decisions are allowed to be endogenous
 56 David Romer’s R&D model

Capital (K) Knowledge (A) Labor (L)

aK 1 – aL
1 – aK aL

Goods-producing Knowledge-
sector producing sector

s Output 1–s
(Y)
New knowledge

Investment Consumption
(C)

 We assume a Cobb-Douglas CRTS production function for goods:

 1
Y  t    1  a K  K  t    A  t 1  a L  L  t  
 Knowledge is produced according to a Cobb-Douglas that may or may not have CTRS:
 
A  t   B  a K K  t    a L L  t   A  t 


A
 B  aK K   aL L  which means growth rate of A (our old g)
 
o Note that  = 1 
A
depends on the amounts of K and L devoted to research and is constant if those
amounts are constant.
o Replication argument cannot be used to justify CRTS here
 Same knowledge produced by two people is not twice as valuable
 Positive spillovers could yield increasing returns to scale
 Are other discoveries substitutes or complements for the next discovery?
 No depreciation and constant saving rate mean K  t   sY  t 
 Exogenous growth of labor force: L  t   nL  t 

Analysis of R&D Model

 Romer begins with a model in which there is no physical capital ( =  = 0)
o We won’t analyze this model in detail, but note the equations of the model if  =
1 and n = 0 (so L is constant)
 David Romer’s R&D model 57

Y  t   A  t 1  a L  L
 A  t 
 B a L L 


A t 
 Output is proportional to A and the growth rate of A is a constant, so this
model has a constant growth rate of output that is determined by B, aL,
and L (and ).
 Higher R&D productivity, more labor being used in the labs, and a bigger
population all lead to a higher growth rate (not just to a higher, parallel
growth path)
 For the full model (with K), we have two state variables, A and K
o We denote the growth rates of A and K by gA and gK
 Dynamics of K
K  t   sY  t    s 1  a k  1  a L   K  t  A  t  L  t 
 1  1 1
 
o The term in brackets is a constant (over time) that we shall call cK
o The growth rate of K at every moment t is
1
 A t  L t  
K  t 
gK t    cK  
K t   K  t  
*
o We seek a steady state in which K grows at a constant rate gK , so we want to
analyze the change in or growth rate of the growth rate
g K  t 
o  1    g A  n  g K  using our rules for growth rates
gK  t 

g K  t   0 if g K  t   g A  t   n 

 g K  t   0 if g K  t   g A  t   n   g K  0 : gK*  g A*  n

g K  t   0 if g K  t   g A  t   n 
 The g K  0 curve is a line with slope of one and intercept on the gK axis
at n ≥ 0.
 Below the line, g K  0 and above the line g K  0 , so the arrows
point vertically toward the line
 Dynamics of A
A  t 
gA t     BaK a L  K  t  L  t  A  t 
  1

A t 
o The term in brackets is constant over time and called cA
 58 David Romer’s R&D model

o As above, the growth rate of the growth rate at every moment t is

g A  t 
  g K  t   n     1  g A  t 
gA t 
n 1   
g A  t   0 if g K  t     g A  t 
  
n 1    n 1   *
 g A  t   0 if g K  t     g A  t    g A  0 : g K*   gA t 
    
n 1   
g A  t   0 if g K  t     gA t 
  
1
 The g A  0 curve is a line with slope and intercept on the vertical

(gK) axis at –n/ ≤ 0.
 To the left of the line g A  0 and to the right of the line g A  0 , so the
arrows point horizontally toward the line
 Equilibrium dynamics
o The nature of the equilibrium depends crucially on two properties of the
parameters:
 n > 0 vs. n = 0
 This determines whether there is any exogenous source of growth
in the model
 If n > 0 as in the Solow and Ramsey models, sustained growth in
total GDP is possible through exogenous growth in L
  +  = 1 vs.  +  < 1 (or  +  > 1)
 This determines “returns to scale in produced inputs”
 Note that K and A are “produced” in the model
 The production function for goods always has constant returns in
produced inputs because K has exponent  and A has exponent
1–
 The production function for knowledge has returns to scale in the
two produced inputs equal to the sum of their exponents:  + 
 If  +  = 1, then the model can sustain ongoing “endogenous”
growth even if n = 0 because increases in both K and A together
are not subject to diminishing returns
 Dynamics with n > 0 and n = 1
o With n > 0, the g K  0 line intercept is positive and the g A  0 line intercept is
negative
o If n = 0, both lines pass through the origin
o Case I:  +  < 1 (diminishing returns in produced inputs)
David Romer’s R&D model 59

1 
 Slope of g A  0 line is  1 , so it is steeper than the g K  0 line


gK

gK*

0
gA* gA

 Economy converges to unique equilibrium from all points in space

 Solving algebraically, we can show

g A*  n
1    
1   
g K*  n
1    
 Growth here is exogenous in the sense that if n = 0, both K and A stop
growing. (Note that both lines intercept at the origin if n = 0.)
 This case replicates the dynamics of the Solow model with g determined
endogenously as a function of n
o Case II:  +  = 1
1 
 In this case, the slope of the g A  0 line  1 and the two lines are

parallel (or coincident)
 If n > 0, then they are parallel
60 David Romer’s R&D model

gK

0
gA

 The economy will move into the channel between the lines and
then growth in both K and A will accelerate forever.
 Intuitively, a bigger economy means more scientists means more
discoveries means faster growth. As long as n > 0, the exogenous
growth in the labor force leads to accelerating growth.
 If n = 0, then the two lines coincide

gK

0
gA

 Economy converged to the line and on the line, both growth rates
are constant and equal (because the line has slope of one)
 Microeconomics of R&D 61

Because gK  gA , K/A is constant in the steady state

* *

o There is a unique K/A* that will sustain equal growth in
K and A and a unique common growth rate g* that is
consistent with that K/A*
o You will work out the algebra in Problem 3.5.
 Example of this case: Suppose that B↑ so that cA increases.
o This raises gA and moves the economy to a point to the
right of the original equilibrium.
o Economy converges back up and to the left to a new
equilibrium that is higher than original.
 Endogenous growth occurs in this case: economy sustains
positive growth even when there is no exogenous source (n = 0)
 Growth rate depends (positively) on s, B, aK, aL, and L
o Case III:  +  > 1
1 
 In this case, the g A  0 line is flatter than the g K  0 because 1

 This case looks like Case II, but the lines are not parallel.
 In this case, we get explosive growth even when n = 0.

Microeconomics of R&D
 The key question that we have dodged in Romer’s R&D model: What determines aK?
o Capital owners must decide whether to build factories or labs
o Economists would assume that the choose the use of their capital that provides
the higher rate of return
 So in equilibrium the amount of capital in the two sectors would have to
balance the marginal rates of return
o Rate of return on factories is straightforward: They produce output that is sold to
earn revenue
 How do labs earn money?
o In the real world, there are lots of funding sources for R&D
 Corporate funding
 Government grants
 Tuition from university students
 Since we don’t model government or university research, we are
interested mostly in corporate-funded research and development
o In our model, knowledge is purely non-rival and non-excludable
 Any discovery is immediately useful to all producers
 62 Model of Learning by Doing

 There is no “appropriability” of knowledge for private benefit

 New knowledge cannot be sold or used profitably
 Why would capital owners put money into labs that earn nothing?
 They wouldn’t, so we would need to build a model of how lab
owners can earn money from R&D in order to pay for the capital
and labor that is used.
 Models of aK
o Corporate R&D is profitable if there is an effective way for the company to
appropriate the knowledge
o This usually occurs by preventing other firms from using the knowledge created
through some kind of “appropriability mechanism”
 May also involve licensing
 Note that either is inefficient, because once created the knowledge is
nonrival and “should” be universally used for free
o Two common appropriability mechanisms are intellectual property rights
(patents) and secrecy
 Both are flawed
 Some kinds of intellectual property are better protected by patents, some
by secrecy, and others are virtually unprotectable
o Effective patent protection or secrecy gives an effective (but usually temporary)
monopoly on the use of the knowledge to the firm doing the R&D
o Two common models for aK are based on this:
 A model of product innovation in which R&D can produce new varieties
of (intermediate) goods on which the innovating firm holds a monopoly
 A model of process innovation in which R&D can advance productive
efficiency of one (intermediate) good (of many) and have a cost
advantage in production until another firm leap-frogs it
 Both models add complexity to Romer’s R&D model because both
require multiple goods in order to have more than one firm
 We study both models in Econ 454

Model of Learning by Doing

 Romer’s short section on learning by doing develops the essence of Paul Romer’s first
(1986) endogenous growth model.
o Kenneth Arrow developed a model in the 1960s based on the idea that a firm’s A
would be increased as it produced output, so A  Y
o Paul Romer’s version of this was slightly different
 Firms’ learning is related to capital accumulation rather than output
 Knowledge is non-appropriable
 Model of Learning by Doing 63

 New knowledge occurs as a by-product of capital investment

 Firms have (some) incentive to invest, so knowledge creation
happens despite pure nonrivalry
 Learning by doing with a constant saving rate
1
Y  t   K  t   A  t  L  t  


o
A  t   BK  t 


1 
Y  t   K  t  B 1 K  t  L t 
 1

o Solving out A yields

1 
K  t   sY  t   sB 1 L  t  K  t 
1

o This model converges, has endogenous growth, or explodes as  < 1,  = 1,  > 1

o Case of  = 1 is the endogenous-growth case
 Let n = 0 so there is no exogenous growth
Y  t    BL K  t   bK  t 
1

K  t   sY  t   sbK  t 
 K  t 
 sb
K t 
 Thus, growth in the capital stock and output is constant at rate sb
 Any increase in saving, in the productivity of learning, or in the labor
force would increase growth
 Ramsey consumers in the learning-by-doing model (not done this way in 4th edition)
o Assume  = 1 and n = 0, so we have the endogenous-growth case
o Aggregate knowledge is proportional to aggregate capital stock (but this is not the
case at the firm level)
 Firms take aggregate knowledge as given and do not consider how their
own investment will add to it because they are small
1
Yi  t   K i  t   A  t  Li  t  


o A  t   BK  t 
Yi  t   B 1 K  t  K i  t  Li  t 
1  1

o The private marginal product of capital is

 1 
Yi  t   Ki t  
K t   r  t  (there is no depreciation)
1 1
 B  
K i  t   Li  t  
Ki
 Each firm sets its so that the private marginal product equals the
Li
economy-wide interest rate r
64 Model of Learning by Doing

Ki
 This means that all firms have the same that is equal to the aggregate
Li
K/L
 Setting Ki/Li = K/L,
1 
 K t  
r  t   B K t 
1 1
 
 L  t  
 1 
 B1 K  t  K t  L t 
1 1

   BL 
1
   BL  t  
1
r
 This rate of return r is constant over time (with n = 0) and depends on
the rate of knowledge accumulation through investment B, , and the
size of the labor force
 Note that the marginal social product of capital (varying K as well as Ki) is
larger than the marginal private product
Yi
  BL 
1
 MSPK   MPPK
K i Ki  K

 This means that individual firms will underinvest in capital

 They do not take into account the positive social externality that
their investment conveys on all firms through increased
knowledge
 This means that the privately generated growth rate will be lower
than the socially optimal growth rate
o Ramsey consumers, as usual, choose a consumption path that satisfies the Euler
C  t  r  t   
equation  .
C t  

C  t 
r     BL 
1

 In this case, r  t   r and   g
C t   
o To satisfy the economy’s budget constraint, Y must grow at the same rate as C, so
the economy grows at g at every instant.
o There are no convergence dynamics: wherever an economy is, it just grows at g
from there. (Poor countries with same parameters do not catch up.)
o Growth rate g depends on parameters of economy: , B, L, , .
o Once again, we have “scale effects” because a larger L means a faster growth
rate.
 If we allow n > 0, then we have both endogenous and exogenous growth
and the growth rate accelerates over time.
 (Paul) Romer model (not worth doing the details) 65

Are scale effects realistic? Some argue no, but Kremer’s argument for
Eurasia, Australia, and Tasmania seems to provide some support.
 In addition, there is much evidence that growth has accelerated over the
centuries (as population has grown).
o Non-optimality: social planner would internalize the knowledge externality and
 BL 
1

use r *   BL   r    BL 
1 1
leading to faster growth at g *  g


(Paul) Romer model (not worth doing the details)

 What’s different about this model?
o We model the incentives for production of knowledge explicitly
o We introduce the “Ethier production function” and the now-ubiquitous model of
a continuum of “intermediate goods”

Human Capital in the Solow Model

 Distinction between knowledge capital and human capital
o Latter is rival and embodied in worker
o Former relates to nonrival ideas that all share (costlessly)
 Model is motivated by the dominant question: “Why are some countries richer than
others?”
o Solow model says differences in k
 Not plausible (as Romer shows late in Ch 1)
o Mankiw, Romer, & Weil: differences in physical and human capital
 They argue this is plausible; others disagree
o Differences in A
 Why would technology be different across countries?
 Barriers (legal and otherwise) to adoption
 Non-applicability of advanced technologies in poor countries (climate,
unreliable physical infrastructure, etc.)
o Differences in “social infrastructure”
 We’ll have more to say about this soon
 How to incorporate human capital into model?
o Many alternative ways; Romer does one (and others in problems 4.8 and 4.9)
o How does economy “produce” human capital?
 Process of education or training has two major costs: teachers’ time (for
which they are paid) and students’ time (for which they are not paid)
 66 Human Capital in the Solow Model

 Can use a two-sector model with a production function for education

using labor (teachers) and capital (schools) like the one for knowledge in
the R&D model
 Can just deduct some amount of a conglomerate “output” as being
education in a one-sector model (like some output is physical capital
rather than consumption). This is Romer’s 4.8.
 Can model the process as holding people out of the labor force during an
education period. This is Romer’s Section 4.1.
 This doesn’t model the cost of teachers and schools.
 Note that forgone earnings may be higher than teacher/school
costs at most schools (if maybe not at Reed)

Simple human-capital model setup

 Let H  t   L  t  G  E  be the amount of human capital, which is the number of workers
L  t  times the amount of human capital per worker G  E  , where E is the average
education level of current workers.
o G E   0
o G  E   e E is a commonly used functional form
o We assume that in a steady state with education level E, people live T years,
going to school for E years and working for T – E years.
o In general (but not in this model), human capital includes not just education but
training, health and other “acquired” characteristics that affect labor
productivity.
1
Y  t   K  t   A  t  H  t  



 K  t   sY  t   K  t 
 A  t   gA  t 
 L  t   nL  t 

Solving the model

 This model looks (and behaves) similarly to Solow model
K K
 Define k  
AH ALG  E 

 k  t   sf  k  t     n  g    k  t 
1
 s 1
 k  0  k  k *   
n g 
 Human Capital in the Solow Model 67

 How will a change in E affect the steady-state growth path?

o Effects of E↑ (or G↑) on K and Y are equivalent to increase in L
o Economy moves to higher, parallel steady-state path
o Level effect, but no growth effect
o Y and Y/L are higher in steady-state
 But the important variable (living standards) here is Y/N, where N is total population
* *
 Y t    L t  
o    y * A  t  G  E    on the steady-state path
 N  t    N  t  
 Increase in E does not affect y* or A(t)
 Increase in E raises G(E)
 Increase in E lowers L/N because more people are in school and fewer in
the labor force
 What will be the net effect?
o What is L/N?
 It seems like it should be T  E  /T since that is the ratio of working
years to total life years for each individual
 That is correct if n = 0
 If the population is growing, then the cohort in education is larger than
the cohort that is working.
 Romer (and Coursebook) shows that in steady state
L t  e  nE  e  nT

N t  1  e  nT

L / N 
 It is intuitively clear (and mathematically easy) that 0
E

Dynamics of increase in E
 Initial effect lowers Y because fewer people in labor force but no immediate increase in
the education of those who are working
 In steady state, the two effects noted above are in conflict and we don’t know which will
dominate
 Y / N   Y / N   Y / N    L / N 
o  
E E  L / N  E
o The first term depends mostly on G   E  and the second is negative.
o If G   E  is large, then Y/N is likely to rise with an increase in E
68 Human Capital in the Solow Model

o This makes intuitive sense: if education is highly productive it will raise per-
capita income; if it is not, then it drains people who could be working into useless
education.