Entrepreneurship, Knowledge, and the Industrial Revolution

Mustafa Aykut Attar

A dissertation submitted to the faculty of the University of North Carolina at Chapel

Hill in partial fulfillment of the requirements for the degree of Doctor of Philosophy in
the Department of Economics.

Chapel Hill
2011

Approved by:

Lutz A. Hendricks

Neville R. Francis

Richard T. Froyen

Pietro F. Peretto

John J. Seater
 c 2011
Mustafa Aykut Attar
ALL RIGHTS RESERVED

ii
 Abstract

Mustafa Aykut Attar: Entrepreneurship, Knowledge, and the Industrial Revolution

(Under the direction of Lutz A. Hendricks)

This dissertation constructs and studies a simple unified growth model that explains the

timing of the industrial revolution through entrepreneurship and entrepreneurs’ previ-

ously unexplored role in the accumulation of useful knowledge.
Three premises are responsible for the main results: First, inventions and discoveries,
i.e. two forms of useful knowledge, are differentiated such that a larger stock of discov-
eries implies a higher level of inventive productivity. Second, entrepreneurs own and

manage the firms operating in the innovative sector of the economy, and they thus may
find it optimal to spend some of their scarce time endowment to inventive activity by de-
creasing the time allocated to routine management otherwise. Third, the stock of useful
discoveries expands in time through the process of collective discovery. Entrepreneurs,

during their lifetime, serendipitously perceive new useful discoveries and share what they
discover with each other in their common social environment.
Two key results are that (i) the optimal level of inventive effort by entrepreneurs is
zero if the stock of useful discoveries is sufficiently small, and (ii) an industrial revolution,

i.e. an endogenous switch from zero to positive inventive effort, is an inevitable outcome
of the process of collective discovery even though it might be delayed for long epochs of
stagnation. Population growth and structural transformation, i.e. two well-documented
aspects of the transition from stagnation to growth, are not only affected by technological
progress as usual but also determine how fast the economy moves towards its invention

threshold.
Calibrated to match some key data moments of England’s economic development

iii
during the last 350 years, the model performs reasonably well in explaining the main
patterns of the transition from stagnation to growth, i.e. the demographic transition,
urbanization, industrialization and the acceleration of technological progress. Counter-
factual experiments show that even small deviations from the benchmark model may

create large effects on the timing of the industrial revolution.

iv
To all my teachers

v
 Acknowledgments

Several people other than my always supportive parents and my brother contributed to
this dissertation, and I would like to express my appreciation to them.

My advisor Lutz Hendricks contributed to this dissertation, firstly, by teaching me

how to design a research agenda and, secondly, by providing comments and suggestions

that made me gain new and important insights about the contents and the scope of my
dissertation. I most possibly benefited from the tacit component of his knowledge on the
way of becoming a research economist.

Other members of the dissertation committee were very friendly and helpful in all
stages of my graduate studies: Neville Francis and Richard Froyen provided valuable

comments that helped me realize some aspects of the dissertation that need to be ad-
dressed further. Pietro Peretto’s research-oriented course on economic growth inspired
me more than anything else. I started thinking about the role of Schumpeterian notions
for unified growth theory after taking that course, and his comments and suggestions

played a key role in finalizing this dissertation. Equally valuable were the comments and
suggestions by John Seater, and I received valuable feedback at the Triangle Dynamic
Macroeconomics workshop organized by him.

Oksana Leukhina shared the data collected by her and her co-author Michael Bar.
Her support saved a good deal amount of time for me, and the data in turn proved to be
very useful.

I benefited from conversations with Aykut Kibritçioğlu and Aydın Ördek on the

dissertation topic, and Michael Salemi provided a very valuable technical comment on an
earlier draft of the dissertation. I appreciate their contributions.

vi
 All my close friends, living now in Chapel Hill or in some other place, were really
supportive. Two of my friends, Volkan Selçuk and Onur Yeni, played a key role and
assisted me financially for my last semester in Chapel Hill. The last but surely not the
least, my girlfriend Günnur’s unconditional support was very important at all stages of

my doctoral research. She kept me alive in times I felt alone and depressed.

vii
 Table of Contents

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii

Chapters

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 The Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1 From Stagnation to Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Three Fundamental Premises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 The Model Economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.1 Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.1.1 Demographic Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.1.2 Endowments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.1.3 Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.1.4 Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.2 Occupations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.2.1 Entrepreneurs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.2.2 Workers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.3 Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.4 Production Sectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

viii
 3.4.1 Modern Sector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4.2 Traditional Sector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5 Decision Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.5.1 Workers’ Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.5.2 Entrepreneurs’ Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3.5.3 Occupational Choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.6 Market Clearing Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

4 Static, Dynamic, and Asymptotic Equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.1 Static General Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.1.1 Ex Ante Symmetry and Ex Post Heterogeneity . . . . . . . . . . . . . . . . . . . 34

4.1.2 The Invention Threshold and the Industrial Revolution . . . . . . . . . . . 35
4.1.3 Productivity, Wage, and Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.1.4 The Structural Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4.1.5 Fertility and the Demographic Transition . . . . . . . . . . . . . . . . . . . . . . . . 39

4.1.6 The Supply of Entrepreneurship . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

4.1.7 Output per worker and Output per capita . . . . . . . . . . . . . . . . . . . . . . . . 41

4.2 Dynamic General Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

4.3 The Asymptotic Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.3.1 Two Preliminary Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4.3.2 The Global Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4.3.3 The Asymptotic Rates of (Economic) Growth . . . . . . . . . . . . . . . . . . . . 46

4.4 The Absence of the Weak and the Strong Scale Effects . . . . . . . . . . . . . . . . . . . . 48
4.5 The Equilibrium Path: From Stagnation to Growth . . . . . . . . . . . . . . . . . . . . . . 49

5 The Timing of the Industrial Revolution: Some Tentative Results . . . . . . . . . . . . . . 52

5.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5.1.1 First-Order and Second-Order Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5.1.2 Threshold vs. Growth Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

ix
 5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

6 The Quantitative Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

6.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.1.1 Time, Generational Growth Rates, and the Initial Period. . . . . . . . . . 60
6.1.2 The Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.1.3 Parameterizing the ξ (K) Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

6.1.4 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.2.1 Some Simulations for the Calibration Period 1650-2000 . . . . . . . . . . . 74
6.2.2 The Equilibrium Path over the Long-Run . . . . . . . . . . . . . . . . . . . . . . . . 76
6.2.3 Counter-factual Experiments and the Timing Question . . . . . . . . . . . 79

7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

7.1 The Industrial Revolution: Break or Continuity? . . . . . . . . . . . . . . . . . . . . . . . . . 82

7.2 Serendipitous Inventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7.3 Managers vs. Inventors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.4 The Fishing Out Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
7.5 The Scale Effect in the Process of Collective Discovery . . . . . . . . . . . . . . . . . . . 88

7.6 Science and Scientists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

7.7 Endogenous Survival and Longevity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

8 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

Appendix A Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

Appendix B The Conditional Dynamical System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

x
 List of Tables

Table 2.1 Two Forms of Useful Knowledge in Mokyr’s (2002) Theory . . . . . . . . . . . 10

Table 2.2 The Usefulness of Knowledge: A Simple Framework . . . . . . . . . . . . . . . . . . 11

Table 4.1 Equilibrium Regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

Table 5.1 The Timing of the Industrial Revolution: Case 1 (nt incr. w/ W t ) . . . . . 56
Table 5.2 The Timing of the Industrial Revolution: Case 2 (nt decr. w/ W t ) . . . . . 57

Table 6.1 The Regression Results for the Exogenous Variables . . . . . . . . . . . . . . . . . . . 64

Table 6.2 The Benchmark Calibration Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Table 6.3 The Timing Effects of Model Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

xi
 List of Figures

Figure 4.1 The Arrival Rate as a Function of the Stock of Discoveries . . . . . . . . . . . . 36

Figure 4.2 The Labor Share of the Traditional Sector as a Function of Real Wage . . 39
Figure 4.3 Net Fertility as a Function of Real Wage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Figure 4.4 The Process of Collective Discovery for t < ∞ and for t → ∞ . . . . . . . . . 47

Figure 6.1 Real Income, Real Wage, and Population in England: 1200-1860 . . . . . . . 60

Figure 6.2 Real NNI per capita and Population: 1650-2000 . . . . . . . . . . . . . . . . . . . . . . . 61

Figure 6.3 The Decline of the Traditional Sector: 1650-1975 . . . . . . . . . . . . . . . . . . . . . . 62
Figure 6.4 Survival Probability and Adult Longevity: Data vs. Fitted Values . . . . . . 65
Figure 6.5 The Measures of Gross Fertility in England: 1650-2000 . . . . . . . . . . . . . . . . 66

Figure 6.6 Model vs. Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

Figure 6.7 The Invention Threshold and the Collective Discovery . . . . . . . . . . . . . . . . 74
Figure 6.8 Some Equilibrium Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Figure 6.9 The Long-Run Evolution of the Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . 76

Figure 6.10 Long-Run Gross Growth Rates (Smoothed) . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

Figure 6.11 The Process of Collective Discovery over the Long-Run . . . . . . . . . . . . . . . 78
Figure 6.12 The Experimental Specifications of Exogenous Variables . . . . . . . . . . . . . . 79

Figure 7.1 Patented Process Innovations in England, 1661-1850 . . . . . . . . . . . . . . . . . . . 83

Figure B.1 The X X t Set and the Dynamics of X t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

Figure B.2 The q q t Locus and the Dynamics of q t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

Figure B.3 The Convergence to the Asymptotic Equilibrium . . . . . . . . . . . . . . . . . . . . . 106
Figure B.4 A Quasi-Static Equilibrium without Productivity Growth . . . . . . . . . . . . . 107

xii
Whether we define the entrepreneur as an "innovator" or in any other
way, there remains the task to see how the chosen definition works out in
practice as applied to historical materials. In fact it might be argued that
the historical investigation holds logical priority and that our definitions
of entrepreneur, entrepreneurial function, enterprise, and so on can only
grow out of it a posteriori. Personally, I believe that there is an incessant give
and take between historical and theoretical analysis and that, though for the
investigation of individual questions it may be necessary to sail for a time
on one tack only, yet on principle the two should never lose sight of each
other. In consequence we might formulate our task as an attempt to write a
comprehensive history of entrepreneurship.

Joseph A. Schumpeter
From his 1949 address at the Research Center in Entrepreneurial History
Cited by Lazonick (1991, pp. 271-272)

xiii
 Chapter 1

Introduction

Why did the Industrial Revolution start when it did? Why not earlier or later? Why was
the stagnation of living standards so prolonged? Which factors did keep today’s developed
societies and others in a quasi-trap of poverty for several millennia?

In the last decade, many growth theorists have returned to this timing question with
the methodology of Galor and Weil’s (2000) Unified Growth Theory: A unified model is
the one that not only features stagnation and growth equilibria but also accounts for the
factors that trigger and govern the gradual transition from the former to the latter.

This dissertation constructs and studies a simple unified growth model that explains
the timing of the Industrial Revolution through entrepreneurship and entrepreneurs’ role
for the accumulation of useful knowledge. The growth of living standards in the model is

due to new inventions created by entrepreneurs who behave very much like Schumpeter’s
(1934) "entrepreneur-inventor"s, and the start of the industrial revolution is an endoge-
nously occurring switch from an equilibrium regime of zero inventive effort to that of
positive inventive effort.

What motivates the emphasis on entrepreneurial invention, in addition to the Schum-

peterian view of the first Industrial Revolution suggested by Solo (1951), Baumol (1990),
Murphy et al. (1991) and Mokyr (2010), is Meisenzahl and Mokyr’s (forthcoming) proso-

pographical evidence on 759 British inventors born between 1660 and 1830. Among 598
inventors with a known business ownership status, only 88 inventors (around 15%) were
employed as non-managers, and 467 of them (around 78%) were business owners. The
latter statistic suggests that understanding the role of inventors who were incentivized by
profit motive during the Industrial Revolution may be of prime importance, and this dis-
sertation develops a simple model that makes entrepreneurs’ role explicit unlike existing

unified growth models.

Three premises are responsible for the main results: First, building on Mokyr’s
(2002) theory of useful knowledge, inventions and discoveries are differentiated such that,
for a given level of effort directed to inventive activity, it is less likely to be success-

ful in achieving a given number of inventions if the number of available discoveries is

smaller. By this premise, recently exploited by O’Rourke et al. (2008) and Strulik (2009)
as well, the model endogenizes the productivity term of a standard invention technol-
ogy that exhibits constant-returns-to-scale with respect to its rival labor input. Second,
entrepreneurs in the model establish and manage their firms in a perfectly competitive

sector that produces the single consumption good in the economy. That entrepreneurs
appropriate positive profit by managing their own firms implies that it may be optimal
for entrepreneurs to allocate some of their scarce time endowment to inventive activity
while decreasing the time spent on routine management, hence the term entrepreneurial

invention as in Grossmann (2009). Finally, the stock of useful discoveries expands in time
through the process of collective discovery. Entrepreneurs, during their lifetime, serendip-
itously perceive new useful discoveries, i.e. new knowledge components about natural
phenomena underlying the production processes but not being themselves inventions,

and share what they discover with each other in their common social environment. This
premise is a way to formalize, albeit imperfectly, what Mokyr (2002) calls industrial en-
lightenment, and it is motivated, among others, by Jacob (1997), Bekar and Lipsey (2004)
and Landes (2006) who emphasize the creation and the diffusion of useful knowledge
among British/European entrepreneurs and capitalists.

Two key results follow from these three premises. First, there exists an inven-
tion threshold: If the stock of useful discoveries is sufficiently small, given (exogenous)

2
longevity, the optimal level of inventive effort by entrepreneurs is zero. Second, the
endogenous mass of entrepreneurs through collective discovery determines how fast the
economy moves to its invention threshold. Provided that an industrial revolution is possi-
ble, it is an inevitable outcome of the process of collective discovery even though it might

be delayed for long epochs of stagnation. The question is thus what factors explain the
mass of entrepreneurs.

To answer this in a parsimonious way within a unified growth framework, the model
incorporates two well-documented aspects of the transition from stagnation to growth,

i.e. population growth and structural transformation, since these determine the mass of
entrepreneurs through the use of economy-wide resources for different tasks.

The model closely follows Jones (2001), Strulik and Weisdorf (2008) and
de la Croix and Licandro (2009) to deliver the following story of demographic change
in the simplest way: Population growth is endogenous through optimal fertility choice

in a two-generation framework. Given exogenous patterns of child survival and adult

longevity, the endogenous evolution of economic prosperity governs the dynamics of
a demographic transition. Specifically, a minimum consumption constraint implies an
upper limit of fertility when the economy is sufficiently poor. An increase in incomes,

within this regime of fertility choice, leads to higher fertility. Once the economy becomes
sufficiently rich, however, this income effect vanishes completely. The optimal level of
fertility declines with technological progress since time, becoming more expensive, in-
creases the unit time cost of a born child. Parents’ strict preference towards reproductive

success, i.e. leaving at least one surviving child, implies a constant level of population
for sufficiently high income levels. Since earth as a closed system has a finite carrying
capacity of people in reality, a constant level of population in the long-run is a desired
feature. Yet, contrary to Peretto and Valente’s (2011) model that explicitly considers the
role of natural resource scarcity, the model simply predicts a constant level of population

in the long-run through preferences towards reproductive success. The model also devi-
ates from the quality-quantity trade-off models of Galor and Weil (2000) and Lucas (2002)

3
by non-homothetic preferences that eliminate the income effect on fertility completely
for sufficiently high income levels. This is, of course, not to say that the endogenous
natural resource scarcity and the quality-quantity trade-off are not important for the de-
mographic transition, but the model is greatly simplified by these deviations and still

returns an empirically plausible story of demographic change.

The way in which the model incorporates structural transformation is similar to

that of Hansen and Prescott (2002) and Bar and Leukhina (2010a) such that (i) a single
consumption good is produced with two technologies, i.e. traditional/rural and mod-

ern/urban technologies, and (ii) the traditional technology admits land as an input in
fixed supply. In contrast with Hansen and Prescott (2002) and Bar and Leukhina (2010a),
on the other hand, (i) technological progress in the modern sector is endogenous through
entrepreneurial invention and (ii) a spillover effect from the modern to the traditional
sector endogenizes productivity change in the traditional sector as in Desmet and Parente

(2009) at some period after the industrial revolution. Under an inequality restriction that
governs the growth of relative productivity in the long-run, the traditional sector faces
an eventual demise. The mapping from the model to the data associates the declines of
the labor and the output shares of the traditional sector in the model respectively with

the rises of urbanization and industrialization as in Bar and Leukhina (2010a).

The trajectory of economic development this dissertation predicts, consistent with

the conventional wisdom of unified growth theory, is as follows: Adult longevity, child
survival probability, the stock of useful discoveries, and urbanization rate are all at their

historically minimum levels in the very beginning. Inventive activity is not optimal, and
the modern/urban sector is small. Poverty is severe, limiting net fertility and population
growth. In time, however, the minuscule increases in adult longevity and child survival
probability, if any, minimally increase net fertility, and the stock of discoveries keeps
expanding albeit at a still slow rate due to the small number of entrepreneurs. This pro-

cess continues for sufficiently many periods to eventually make inventive effort optimal.
An industrial revolution starts. Increasing wages imply faster urbanization and higher

4
gross/net fertility first. Once productivity is sufficiently high, however, adult individuals
find it too costly to produce too many children and limit fertility. Urbanization on the
other hand keeps accelerating to make the mass of entrepreneurs even larger and collec-
tive discovery even more productive. Thus, industrialization accelerates. The asymptotic

equilibrium features perpetual growth of labor productivity and output per capita with
a strictly positive asymptotic growth rate and a constant level of population.

The quantitative analysis of the model calibrates the structural parameters and the
initial values of state variables to match the observed patterns of economic development

in England from 1650 to 2000. England is not only the first industrialized economy
but also the one for which a sufficiently large set of data exists. The simulations for the
period from 1650 to 2000 show that the model economy performs reasonably well in
explaining population and output per capita dynamics and the rises of urbanization and
industrialization. The calibrated model is next used for some counter-factual experiments

on the timing of the industrial revolution as in Desmet and Parente (2009). The results,
interpreted only as suggestive and only for the timing of England’s industrial revolution,
indicate that small deviations from the benchmark model may create large timing effects.

The contribution of this dissertation to the literature can be described as follows: First

and the foremost, the dissertation shows that thinking a bit seriously about the produc-
tivity term of an otherwise standard invention technology and bringing the entrepreneur
back to the scene of economic development allow us to understand why purposeful in-
vention may not be optimal for a very long episode of history and why an industrial rev-

olution is inevitable. Entrepreneurs’ dual role for inventions and discoveries in a unified
growth framework, motivated strongly by the anecdotal and prosopographical evidence,
is the previously unexplored mechanism proposed in this dissertation. Second, the model
eliminates both the weak and the strong scale effects of population size as in Grossmann
(2009). The absence of the weak scale effect is important because the role of population

size during the modern growth regime remains ambiguous in most unified growth mod-
els. The strong scale effect, on the other hand, is ruled out as the mass of entrepreneurs

5
plays the role of the horizontal dimension of technological progress of the second-
generation Schumpeterian models of Young (1998), Peretto (1998a), Aghion and Howitt
(1998, Ch. 12) and Dinopoulos and Thompson (1998). Third, due to stochastic invention
and ex post heterogeneity across firms in the modern sector, the model explains, unlike

other unified growth models, (i) why more innovative firms on average are larger and
why the size distribution of innovative firms are skewed; two well-known regularities
most recently reiterated, respectively, by Akcigit and Kerr (2010) and Klette and Kortum
(2004). Finally, although the model does not incorporate human capital in its usual sense,
that it does not crucially require a rise in the demand for embodied skills for the industrial

revolution to start, as that of O’Rourke et al. (2008), answers Clark’s (2005) criticism of
unchanged skill premium during the first Industrial Revolution.
The plan of the dissertation is as follows: The next chapter provides a discussion of
the related literature to more clearly locate the contribution of this dissertation. Chapter

3 introduces the model economy. Chapter 4 defines and analyzes the uniquely existing
static, dynamic and asymptotic equilibria of the model and then characterizes a possible
equilibrium path from a unified growth perspective. Chapter 5 provides some tentative
(analytical) results on the timing of the industrial revolution. Chapter 6 presents a quan-

titative analysis of the model economy. Chapter 7 provides a discussion of some aspects
of the model and manipulates over some possible extensions. Chapter 8 concludes with
some final remarks.

6
 Chapter 2

The Related Literature

This chapter presents a discussion of the related literature in two sections: First, the no-
table contributions to the literature studying the transition from stagnation to growth

are mentioned, and the relation of the dissertation with this literature is specified. Sec-
ond, scholars who have contributed earlier on the three fundamental premises of the
model, i.e. the discovery-invention distinction, entrepreneurial invention, and collective
discovery, are credited, and the contribution of this dissertation is described.

2.1 From Stagnation to Growth

Poverty trap models of Murphy et al. (1989), Azariadis and Drazen (1990), Becker et al.
(1990) and Matsuyama (1991) predict multiple steady-state equilibria, but these models

leave the mechanism that explains the transition from stagnation to growth unexplored.1
An early study that deals explicitly with the transition is of Goodfriend and McDermott
(1995) who argue that there exists a critical mass of population that makes industri-
alization efficient through the market size effect. Tamura (1996) extends the model

of Becker et al. (1990) by a clearer account of transitional dynamics. Building on

1. Each of these models focuses on a different mechanism that generates development thresholds and mul-
tiple equilibria. Murphy et al. (1989) emphasize the role of demand spillovers across industries by for-
malizing the Big Push argument. Azariadis and Drazen (1990) study the externalities in the accumula-
tion of human capital that create threshold effects. Building on the fertility theory of Becker and Barro
(1988), Becker et al. (1990) utilize the quality-quantity trade-off of human capital accumulation to gen-
erate Malthusian stagnation and modern growth as two (stable) steady-state equilibria. Matsuyama
(1991) develops a two-sector model of occupational choice with increasing returns in manufacturing.
Azariadis and Drazen’s (1990) model, Arifovic et al. (1997) endow agents by adaptive
learning that eventually triggers the takeoff. Acemoglu and Zilibotti (1997) explain the
minuscule rates of capital accumulation and economic growth during the early stages of
development by the absence of complete financial markets that diversify risk. All these

early models, however, fail to account for the demographic transition.

Galor and Weil’s (2000) model, for the first time, successfully explains not only the
prolonged Malthusian stagnation and the endogenous occurrence of an industrial revo-
lution but also the demographic transition. The authors show that, once the Malthu-

sian constraint of subsistence consumption and the quality-quantity trade-off of human

capital accumulation in the fashion of, e.g., Becker et al. (1990) are conveniently inte-
grated with the skill-biased technological change and with the slow pace of technology-
population interaction as in Boserup (1965) and in Kremer (1993), the desired result fol-
lows; a gradual and endogenously occurring phase transition from stagnation to growth

that does not require a major shock.

Notable contributions to the literature include that of Hansen and Prescott (2002)
who study the decline of the land-based sector and the rise of industrialization within
a neoclassical framework with exogenous technological progress. Lucas (2002) merges

this framework with endogenous fertility choice and human capital accumulation. The
emphasis on the demographic transition as a result of endogenous fertility choice is main-
tained by Jones (2001) who extends the analysis, for the first time, by a very simple
formulation of incentives to innovate. The decline of agriculture is studied (i) within

the framework of Hansen and Prescott (2002) but with endogenous growth in manu-
facturing sector by Kögel and Prskawetz (2001) and (ii) within Galor and Weil’s (2000)
framework by Tamura (2002). The analysis of the demographic transition is extended
further by Hazan and Berdugo (2002) and Doepke (2004) who introduce child labor, by
Boldrin and Jones (2002) who revisit the old-age security hypothesis with game theoret-

ical notions, and by Lagerlöf (2003) who endogenizes mortality. Galor and Moav (2002)
present an evolutionary unified model of economic growth in which the distribution

8
of the valuations for human capital accumulation across individuals is subject to natural
selection.2

A majority of unified growth models either do not utilize the invention of new
products and new processes as the source of technological progress at all or productiv-
ity change is incorporated in a reduced-form way such that the mechanism of endoge-

nous technological progress remains as a black-box. Motivated partly by this dissatis-

faction and partly by the critiques of some economic historians such as Mokyr (2002),
Crafts (2005), Lipsey et al. (2005) and Clark (2007) who keep stressing the role of inven-
tions for understanding the transition from stagnation to growth, some recent unified

growth models explicitly deal with the incentives leading agents to innovate, be them
individual "consumer/producer"s or firms. O’Rourke et al. (2008) study the transition
from unskilled-labor-biased technological change to the skill-biased technological change.
Desmet and Parente (2009) put the emphasis on the evolution of market competition
that, once sufficiently intensified, activates purposeful innovation by monopolistic firms.

Milionis and Klasing’s (2009) model, featuring "consumer/producer"s, explains why hu-
man capital accumulation eventually leads to purposeful invention. In Bar and Leukhina
(2010b), individuals optimally choose the time they spend on innovative activities while
the main emphasis is directed to the effects of mortality on knowledge transmission. The

last but not the least, Madsen et al. (2010) show that the last four hundred years of British
economic growth can best be understood as a story of endogenous technological progress.

The model constructed in this dissertation contributes to this second strand of the
unified growth literature that endogenizes technological progress with an explicit inven-
tion technology and profit-seeking agents. The model is truly unified in the sense that

the gradual phase transition from stagnation to growth occurs endogenously. Most im-
portantly, the new mechanism proposed features entrepreneurs’ dual role for inventions
and discoveries.

2. Unified growth literature is by now large and diversified, and a complete review is not central for the
purposes of this dissertation. See Galor (2005, 2010) for two useful surveys of the literature.

9
 Table 2.1: Two Forms of Useful Knowledge in Mokyr’s (2002) Theory

Propositional Knowledge Prescriptive Knowledge

Epistemic Technical
Laws and Principles Blueprints and Recipes
What-Knowledge How-Knowledge
Discoveries Inventions

2.2 Three Fundamental Premises

The conceptual framework of useful knowledge that this dissertation builds upon, with

discoveries and inventions being distinct knowledge forms, is due to Mokyr (2002). In
his theory, discoveries are propositional forms of knowledge that do not have direct tech-
nological applications. Discoveries are laws and principles that answer "What?" ques-
tions about natural phenomena underlying the production processes. Inventions, in
contrast, are prescriptive in the sense that they provide answers to "How?" questions;

inventions take the forms of blueprints and recipes. Table 2.1 presents this conceptual
framework, and Table 2.2 remarks the two roles of useful knowledge studied in this
dissertation. Other than Mokyr (2002), the role of the discovery-invention distinction
and the usefulness of discoveries for inventive activity have been emphasized by Landes

(1969), Rosenberg (1974), Nelson (1982) and Easterlin (1995) in explaining the history
of technological progress. In one context, Weitzman (1998, p. 345) has suggested that
knowledge accumulation has distinct recombinant and productivity aspects. The former
corresponds, in a sense, to the role of discoveries for inventions. With reference to knowl-

edge capital, Lucas (2002, p. 12) has asked "[w]hat can be gained by disaggregating into
two or more knowledge-related variables." The model studied in this dissertation answers
this question by showing that, when the productivity of inventive effort is endogenous
to how large the stock of useful discoveries is, purposeful and costly invention may re-
main suboptimal. The distinction between discoveries and inventions is also emphasized

by Haruyama (2009) in an endogenous growth model with perfectly competitive innova-

tion. Howitt and Mayer-Foulkes (2005), O’Rourke et al. (2008) and Strulik (2009) are the

10
 Table 2.2: The Usefulness of Knowledge: A Simple Framework

Discoveries Inventions
useful in increasing useful in increasing
the productivity of the productivity of
inventive effort worker hours

ones who have incorporated the distinction into the formal analysis of unified growth.
However, the dual role of entrepreneurship for useful knowledge remains unexplored in

a unified model with population growth and structural transformation. This is what this
dissertation attempts to deliver.

At least since Schumpeter (1934), "entrepreneur-inventor"s are leading actors of the

narratives of the Industrial Revolution. Solo (1951), Baumol (1990), Murphy et al. (1991)
and Mokyr (2010), among others, have argued specifically that entrepreneurial invention

was indeed the engine of technological progress during the first Industrial Revolution
long before the rise of modern R & D lab. Peretto (1998b) has emphasized the distinction
between "entrepreneurial invention" and corporate R & D in a second generation Schum-
peterian model. Michelacci (2003) has previously studied the role of entrepreneurial

skills in bringing inventions to markets. From another perspective, Doepke and Zilibotti
(2008) and Galor and Michalopoulos (2009) have studied the role of entrepreneurial traits
for long run economic development. This model, differently from all these, incorpo-
rates both occupational choice and entrepreneurial invention within a unified growth

setting. Two-occupation framework of the model is similar to, and even simpler than,
those of Lucas (1978), Murphy et al. (1991) and Michelacci (2003), and the formulation
of entrepreneurial invention under perfect competition shares similarities with the treat-
ments of Hellwig and Irmen (2001), Grossmann (2009) and Haruyama (2009).

The process of collective discovery by entrepreneurs has been described by Landes

(2006, p. 6) as "the seventeenth-century European mania for tinkering and improving."

Bekar and Lipsey (2004) go further to argue that the diffusion of Newtonian mechanics
among British industrialists was the prime cause of the first Industrial Revolution. A

11
similar argument on the diffusion of scientific culture, again with an emphasis on British
success, is made by Jacob (1997). Kelly (2005) develops a network model that analyzes
this type of collective learning for the industrial revolution. Lucas (2009) also emphasizes
collective learning in a model that differentiates propositional knowledge from produc-

tivity. O’Rourke et al. (2008) and Milionis and Klasing (2009), with environments simi-
lar to that of Galor and Weil (2000), link the accumulation of propositional knowledge to
human capital accumulation respectively through the number of high-skilled individuals
and the individual-level stock of skills. Howitt and Mayer-Foulkes (2005) assume that the
skill level of entrepreneurs is proportional to the average productivity associated with

intermediate inputs of production. The last but not the least, Strulik (2009) suggests that
propositional knowledge grows through learning-by-doing at the firm level. What dif-
ferentiates this paper’s formulation of propositional knowledge is the role of the mass of
entrepreneurs. More entrepreneurs with longer lives create more useful discoveries given

the quality of creating and diffusing these discoveries. This type of scale effect by which
knowledge growth depends not on the mass of entire population but instead on a certain
mass of urban population is emphasized more recently by Crafts and Mills (2009).

12
 Chapter 3

The Model Economy

This chapter first introduces the model environment and then specifies market and own-
ership structures. Next follow the occupational structure and the decision problems
solved by each occupational group. Market clearing conditions at the end close the model.

The model economy features some simplifications that are not uncommon in the uni-
fied growth literature: The economy is closed, there does not exist a political authority,

and there is no physical capital. This is an economy of a single consumption good pro-
duced (i) with a traditional/rural technology to which land is an essential input and (ii)
with a modern/urban technology that is utilized by entrepreneurs.

3.1 Environment

The calendar time of the model economy, denoted by t , is discrete with an infinite hori-

zon: t ∈ N+ .

3.1.1 Demographic Structure

There exist two overlapping generations: Individuals who are adults in period t give birth
to children at the beginning of period t , i.e. at the beginning of their adulthood. Their
surviving children become adults in period t + 1.
Fertility

Reproduction is asexual, and b t ∈ R++ denotes the total number of children a generic
adult in period t optimally chooses to give birth to, i.e. the gross fertility per adult.
Note that b t , not being discrete in the model, represents the average gross fertility among
period-t adults, and the total fertility rate of the economy is equal to 2b t under the as-
sumption that the sex ratio of population is 1/2.

Survival Probability

The survival of a child born at the beginning of period t is a Bernoulli event with the
survival probability st ∈ [0, 1]. For simplicity, (i) st is common across period-t children,
(ii) st is known by period-t adults with certainty, and (iii) the sequence {st }t ∈N+ of survival
probabilities is exogenous. Strongly motivated by actual data from developed economies,

{st }t ∈N+ is assumed to be a non-decreasing sequence satisfying

s0 < 1 and s ⋆ ≡ lim st = 1

t →∞

At t = 0 that corresponds to some distant past, only some of the children born to an
adult survive to adulthood. The distant future, on the other hand, represents an era at

which the health/mortality revolution is complete so that all children survive.

Population Growth

Notice that, given b t and st , the gross growth rate of adult population, i.e. the net fertility
per adult, satisfies nt ≡ st b t . Denote by N t ∈ R++ the adult population in period t and
assume that N0 > 0. The law of motion of N t is

N t +1 = nt N t (3.1)

14
Adult Longevity

Not less important than the number N t of adult individuals is how long they live on
average since labor is an input of the production and the invention technologies.3 The
simplest way of capturing the role of adult longevity within this two generations frame-
work is to assume, as in Hazan and Zoabi (2006), that all period-t adults live a frac-

tion ℓ t ∈ [0, 1] of period t . For simplicity, again, (i) ℓ t is common across period-t
adults, (ii) ℓ t is known by period-t adults with certainty, and (iii) the sequence {ℓ t }t ∈N+
of longevity fractions is exogenous. Also motivated by actual data from developed
economies, {ℓ t }t ∈N+ is assumed to be a non-decreasing sequence satisfying

ℓ0 < 1 and ℓ⋆ ≡ lim ℓ t = 1

t →∞

The reasoning behind the survival probability thus applies to adult longevity: In
some distant past, adult individuals do not live up to the maximum lifespan. When
the health/mortality revolution is complete in distant future, however, they enjoy the

longest life.

3.1.2 Endowments

Labor

Normalizing the length of a period to unity implies that period-t adults have a time en-

dowment of ℓ t units each. This time endowment is the only source of their homogeneous
labor force. Children in contrast do not have a time endowment, and they remain idle
until they become adults next period.

3. Adult longevity plays a key role for the timing of the industrial revolution in this model by affecting
the optimality of inventive effort for any t .

15
Land

Land is a production factor of the traditional technology, and the total land endowment
of the economy is normalized to unity. What fraction of individuals own land and at
what proportions, on the other hand, are of minor importance from an analytical point
of view since, as we shall see below, the model builds upon a very simplified view of what

happens in the sector using the traditional technology as in Galor and Weil (2000) and
Desmet and Parente (2009).

3.1.3 Preferences

A generic period-t adult derives lifetime utility from her consumption C t and net fertility
nt .4 The utility function representing these preferences is

U (C t , nt ) ≡ C t + φ ln(nt ) φ>0 (3.2)

with boundary restrictions

Ct ≥ γ > 0 (3.3)

nt ≥ 1 (3.4)

Strulik and Weisdorf (2008) and de la Croix and Licandro (2009) build on this type of

non-homothetic preferences to understand the various aspects of the demographic tran-

sition. Here, a discussion of the notions embedded in (3.2)-(3.4) is necessary:
The preference parameter φ > 0 determines the strength of parental desire towards
reproduction. Such a parameter is a common element of economic models of demogra-

phy.
The non-homotheticity of preferences is important for two reasons: First, the risk
neutrality of preferences with respect to C t eliminates the direct income effect on fertil-

4. Note that choosing n t is identical to choosing b t since s t is known with certainty.

16
ity. This guarantees fertility to decrease with the cost of reproduction when the econ-
omy is sufficiently rich (C t > γ ). Fertility theories based on homothetic preferences
face difficulties to generate a negative fertility-income relationship (Jones et al., 2011),
and the model assumes away such difficulties, in part, with these non-homothetic pref-

erences. The second reason of adapting risk neutral preferences with respect to C t is
to simplify the decision problem of entrepreneurs. As we shall see below, the decision
toward entrepreneurial invention is in essence an expected utility problem with occupa-
tional choice.

The preference parameter γ > 0 in (3.3) denotes some baseline or "subsistence" level
of adult consumption as in Galor and Weil (2000), Jones (2001) and others; γ is the level
of consumption that partially determines the upper bound of fertility when the economy

is sufficiently poor (C t = γ ). It shall be clear below that (3.3) simply reintroduces a direct
income effect on fertility for a sufficiently poor economy: An adult’s consumption is her
first priority, and she has to adjust her fertility accordingly.5

The inequality in (3.4) represents the parental preference for reproductive success in
transmitting genes to the next generations. What motivates this is the long run stability
of population levels observed in nature. A successful parent is the one who leaves at least
one surviving child.6 (3.4) is a very simple way of introducing reproductive success in a

model of fertility choice, but this suffices to produce the desired property:7 The baseline
level of net fertility is equal to unity as in Jones (2001), and this implies a stabilizing level
of population when the economy is sufficiently rich.

5. In this model, γ represents subsistence not merely in biological terms; Landes (1969, pp. 13-14) and
Voth (2003, p. 224) argue that Western European economies in general and England in particular were
richer than many other economies around the world on the eve of the Industrial Revolution.

6. Recall that all surviving children become fecund at the beginning of their adulthood by construction.

7. de la Croix and Licandro (2009) and Strulik and Weisdorf (2011) incorporate reproductive success, re-
spectively, into continuous and discrete time environments where parents choose not only the number
of children they have but also the likelihood of these children’s survival. This, in a sense, is a biological
version of quantity-quality trade-off. With exogenous (s t , ℓ t ), the model abstracts from this.

17
3.1.4 Technologies

Reproduction

There are two inputs of reproduction: Each child born, whether she survives or not,
requires ρ > 0 units of time to be raised to adulthood and consumes ψ > 0 units of
consumption good provided by her parent.8

Time cost of reproduction is a common theme of a vast majority of fertility theo-

ries, and it plays an essential role in this model of demographic transition. Simply, the
cost of reproduction increases in an economy that records sustained increases in labor

productivity since a unit of time becomes more expensive.

Goods cost of reproduction, albeit not necessary for fertility to decline in advanced

stages of economic development, adds to the consistency of the model in which adults
must consume a minimum amount of γ > 0.

The Modern Technology

The following mention of firms and entrepreneurs is necessary to introduce the remain-

ing technologies of the model with maximum clarity: In any competitive equilibrium of
this model economy, (i) there exists a set [0, E t ] of firms that operate with the modern
technology of production, and (ii) each firm using the modern technology is owned by
an entrepreneur. Hence, the set of entrepreneurs is also [0, E t ] where E t < N t is the

endogenously determined mass of entrepreneurs.9

8. In contrast with existing models featuring a time cost of reproduction, this model does not restrict the
time spent on reproduction to be the parent’s time exclusively. We shall see below that this interpreta-
tion is necessary for a particular simplifying assumption on the cost of reproduction for entrepreneurs.

9. Also note that, since there exists a unique consumption good in this economy, the traditional technol-
ogy might be sufficiently productive, compared to the modern technology, to imply E t = 0 at least
initially. Since the purpose here is to study the role of entrepreneurship, the rest of the analysis im-
plicitly restricts (i) the model’s fixed parameters, (ii) the exogenous (state) variables, and (iii) the initial
values of endogenous state variables such that E t > 0 for all t .

18
Production: Let i ∈ [0, E t ] be a representative firm. The Cobb-Douglas technology of
production is

λ 1−λ
Yi t = Xi t hw i t h mi t
λ ∈ (0, 1) (3.5)

where Yi t denotes output, Xi t > 0 denotes labor productivity associated with worker
hours hw i t , and h mi t denotes management hours allocated to production.

Recall that the labor endowment of (adult) individuals is homogeneous. Accordingly,

what differentiates worker and management hours in (3.5) is only the nature of the tasks

in the question: Two distinct tasks are required to produce the good. Workers are the ones
who actually produce the good in its finalized form with their eye-hand coordination, and
managers are the ones who tell workers what to do and how to do it.

The knowledge content of management is represented by Xi t which is a measure of

prescriptive knowledge; Xi t identifies the "quality" of the production process.

Productivity Change: What cause the labor productivity of the modern technology to
grow in time are new inventions created through costly inventive projects. The invention
projects, requiring research hours directed to invention, generate a stochastic number of

inventions, and each new invention increases a baseline level of productivity by some
fixed factor.

Suppose, as in Desmet and Parente (2009) and many others, that the modern sector
firms in period t have access to the average productivity X t ∈ R++ attained by the firms
of the previous generation. The term access here refers to the intergenerational diffusion

of prescriptive knowledge embedded in X t . Simply define this average as

Z Et −1
Xt ≡ E t−1
−1
X j t −1 d j (3.6)
0

where j indexes period-(t − 1) firms.

19
 Define now firm i’s operating productivity as

X i t ≡ σ zi t X t (3.7)

where σ > 1 is the stepsize of inventions and zi t ∈ N+ is the stochastic number of inven-
tions satisfying
zi t ∼ Pois ai t (3.8)

In (3.8), ai t ∈ R+ denotes the arrival rate of inventions and satisfies

θ>0 (3.9)


ai t = θξ K t h r i t

where h r i t denotes the amount of hours allocated to research by firm i. This invention

technology features constant returns to scale with respect to rival labor input h r i t .

The novelty here is the term θξ K t , i.e. the level of research productivity per unit

of inventive effort, where K t ∈ R++ denotes the number of costlessly accessible useful

discoveries to be utilized in the process of inventing. ξ K t , assumed to be continuously

differentiable, is the intrinsic component of research productivity explained by epistemic

foundations; θ > 0 is the extrinsic productivity of invention.

Three restrictions on ξ K t are central to the main results of this paper, and it is

more convenient for the sake of the discussion to formally state them first:

ξ ′ (K) > 0 for all K < ∞ ξ (0) = 0 lim ξ (K) = 1 (3.10)

K→∞

The first restriction here represents the conjecture that it is more likely to be success-

ful in invention if more useful discoveries are available. In light of the earlier discussion
about the role of propositional knowledge, this is a plausible way to restrict the epistemic

20
function ξ K t .10

The second restriction imposes that discoveries are essential inputs of invention

projects. In Mokyr’s (2002, pp. 13-14) words, "[t]he likelihood that a laptop computer
would be developed in a society with no knowledge of computer science, advanced elec-
tronics, materials science, and whatever else is involved is nil."

The third restriction originates from the conjecture that the productivity of research
per unit of inventive effort should have an upper limit when K t → ∞ as in, e.g.,
Weitzman (1998): The arrival rate is a growth factor of productivity, and this has to

have an upper limit naturally imposed by (neuro)physiological capabilities of humans;

the arrival rate should be bounded above for any level of inventive effort that is less than
infinity. That K t goes to infinity means, on the other hand, that every single knowable
thing about the natural phenomena underlying the production processes is known. This

is the ultimate enlightenment that eliminates what causes a unit of inventive effort to be
less productive than its full potential of θ, i.e. the narrowness of the epistemic bases for
invention. An inventor, knowing practically everything about the natural phenomena,
does not have to spend her time with trying to realize which certain discoveries are use-
ful. She simply generates an expected number of inventions with constant (maximum)

productivity θ as in many other endogenous technology models.

The Traditional Technology

Production: The Cobb-Douglas production function that determines the volume YT t

of output produced with the traditional technology is

€ Šη 1−η
YT t = X̃T t HT t LT t η ∈ (0, 1) (3.11)

10. Clearly, the invention technology (3.9) does not incorporate the fishing out effect: Research produc-
tivity per unit of inventive effort does not decrease with X t . Chapter 7 discusses an alternative postu-
lation where research productivity increases with K t and decreases with X t .

21
where X̃T t is labor productivity, HT t is the total amount of hours allocated to production,
and LT t is land input.

Productivity Change: Since the main purpose is not to develop a theory of traditional
technology’s labor productivity growth, a very simple law of motion for X̃T t is adapted
following Desmet and Parente (2009):

 
 ζ1
X̃T t +1  X t +1 
= max ζ0 ,   ζ0 > 1, ζ1 > 0 (3.12)
X̃T t  X t


In (3.12), ζ0 > 0 denotes an exogenous (gross) growth rate of X̃T t . The second ar-

gument of the maximum function imposes that X̃T t grows faster than ζ0 if the spillover
effect from the modern technology is high enough. (3.12) thus captures initially slow and
stable and then accelerating rates of productivity growth of the land-based technologies.

Collective Discovery

The last technology to be specified is of the stock K t of useful discoveries. In its generic
form, what creates new discoveries is simply time. The conjecture, as introduced earlier,
is that entrepreneurs, owning the firms using the modern technology, collectively dis-
cover new pieces of propositional knowledge during their lifetime. They not only create

new knowledge in this serendipitous way individually but also share what they create
with each other in their common environment, e.g. in coffeehouses. This is a network
effect that is consistent with the common knowledge characterization of useful discover-
ies.

The simplest way to formalize this is a linear knowledge production function of the
form
K t +1 − K t = ωℓ t E t ω>0 (3.13)

where ℓ t E t denotes the total lifetime of all entrepreneurs that, interpreted as a single

22
variable, is the input of the collective discovery process and ω > 0 represents the quality
of environment for creating and sharing useful discoveries. This constant thus represents
geographical, cultural and social determinants of collective discovery process.11

3.2 Occupations

Recall that the set of entrepreneurs is [0, E t ] in equilibrium, and i indexes entrepreneurs.
The remaining mass N t − E t of adult individuals forms the set of workers.

Equilibria of this model build upon two decision problems solved respectively by
entrepreneurs and workers. The purpose of this and the following sections is to derive
these decision problems.
The problems involve fertility choice for both occupations and an additional decision

regarding the inventive activity for entrepreneurs. These decisions are finalized in the
sense that they embed other decisions these individuals have to take optimally. That is,
if the occupation is chosen optimally and the decision problems are solved, both at the
beginning of adulthood, other actions follow contingently.

To proceed with maximum clarity, it is necessary at this stage to briefly discuss what
entrepreneurs and workers do.

3.2.1 Entrepreneurs

An entrepreneur, as noted earlier, establishes a firm that uses the modern technology.
There are no establishment costs involved. Once an entrepreneur establishes the firm,

she becomes the business owner until the end of her life. The firm dies with its en-
trepreneur, and each generation raises new entrepreneurs who establish and own new
firms using the modern technology.
As the business owner, an entrepreneur purchases an optimal level of worker hours.
The other necessary input, i.e. management, is provided by the entrepreneur herself.

11. Note that the linearity with respect to ℓ t E t is not a crucial assumption of the model. The qualitative
nature of results does not change as long as K t +1 − K t is an increasing function of ℓ t E t .

23
This creates an incentive for the entrepreneur to spend resources for inventive activities.

An entrepreneur, as any adult individual, also chooses her fertility. The key simpli-
fying assumption here is that entrepreneurs hire workers to take care of their children.12

From a technical point of view, this simplifying assumption is necessary to separate the
choice of fertility from the decision regarding the inventive activity for an entrepreneur.

3.2.2 Workers

Those who choose to become workers supply hours for three activities:

First, some workers are employed by entrepreneurs to work in their firms using the
modern technology.

Second, some workers, again being employed by entrepreneurs, take care of en-
trepreneurs’ children at home.

Finally, some workers choose to work in the firms that use the traditional technology.

3.3 Markets

All sellers and all buyers in any market of this economy are price-takers. This is a model
of perfect competition.

Two things are traded in markets. First, the consumption good is demanded by all

adult individuals in the economy and produced/sold by firms that use either the tradi-
tional or the modern technology. The consumption good is the numéraire of the econ-
omy. Second, a worker hour is traded at wage W t > 0.

12. One justification for this assumption, despite its sexist tone, is this: In the British inventors data set of
Meisenzahl and Mokyr (forthcoming), we observe only one woman out of 759 individuals. If it is not
entirely misleading to imagine that mothers take care of their children and a business is run by one
person only, then the cost of fertility for a family where the husband chooses to be an entrepreneur
must be foregone worker earnings.

24
3.4 Production Sectors

3.4.1 Modern Sector

Entrepreneurial invention occurs, if it is optimal, in the modern sector and generates

positive economic growth through productivity increases. Then, the question is what
creates an incentive to invent for each entrepreneur under perfect competition. That
each entrepreneur manages her firm by providing the manager hours herself answers
this question: Management is basically the quasi-fixed factor of production that implies

positive profit for the entrepreneur. Assuming that the invention technology is costlessly
accessible, it shall be the case that the entrepreneur may find it optimal to allocate some
of her scarce time endowment to inventive activity.

Optimal Ex Post Profit

The first task in characterizing what happens in the modern sector is to derive the optimal
ex post profit maximized through the optimal choice of the demand for worker hours.

Ignoring the complications regarding the timing of events within a period and follow-
ing, e.g., Aghion and Howitt (2009), assume that an entrepreneur can choose the level of

optimal demand for worker hours hw i t for given (Xi t , h mi t , W t ) to maximize profit. The
profit function is defined as

λ 1−λ
Π hw i t , Xi t , h mi t , W t ≡ Xi t hw i t h mi − W t hw i t (3.14)


t

The optimal hw i t and Π hw i t , Xi t , h mi t , W t thus satisfy

1 λ
λ 1−λ Xi 1−λ
t
h mi t
hw i t ≡ arg max Π h, Xi t , h mi t , W t = (3.15)


1
h 1−λ
Wt

‚ Œ λ
λ Xi t 1−λ

Πi t = (1 − λ)λ 1−λ h mi t (3.16)

Wt

25
Entrepreneurial Invention

Replace Xi t with σ zi t X t . The invention technology, via (3.8) and (3.9), then implies the
expected profit as in

 λ

exp −ai t

   1−λ
∞ az z
X λ σ Xt
 it
 
EΠi t ≡  (1 − λ)λ 1−λ   h mi t  (3.17)
z=0 z!  W t


The term in the first brackets denotes the Poisson probability of generating z inven-

tions given ai t . The term in the second brackets is the level of optimal ex post profit when
entrepreneur generates z inventions given ai t .

What does not allow entrepreneur i to be able to spend an infinite amount of re-
sources to inventive activity is her time constraint: h mi t + h r i t ≤ ℓ t . Since EΠi t is strictly

increasing in h mi t , this constraint holds with strict equality in equilibrium:

h mi t + h r i t = ℓ t (3.18)

Together with (3.9), (3.18) allows us to rewrite the expected profit as

 λ

exp −ai t

   1−λ !
∞ az σ z
X ai t
X λ t
 it (3.17′ )
 
EΠi t =  (1 − λ)λ 1−λ   ℓt −

z=0 z!  W t θξ K t 

Lemma 1: The expected profit in (3.17′ ) can be rewritten as

 Γ !
Xt ai t
EΠi t = exp Σai t Λ  ℓt − (3.17′′ )




Wt θξ K t

where
λ λ λ
Γ≡ Λ ≡ (1 − λ)λ 1−λ Σ ≡ σ 1−λ − 1
1−λ

26
are defined for notational ease.
Proof — See Appendix A. „

(3.17′′ ) now identifies the cost of and the return to inventive activity. The first term,

exp Σai t , increases the expected profit by some factor greater than one depending on

the size σ of invention and the productivity elasticity λ of output. The last term in the
last parentheses, ai t /θξ K t , is the time cost of inventing with an expected number ai t

of inventions. Clearly, (3.17′′ ) specifies the deterministic level of profit when no inventive

activity is undertaken (ai t = 0). Xi t is equal to X t in this case.

Production

The production at the firm level is implied by (3.5) given (hw i t , Xi t , h mi t ) where hw i t
satisfies (3.15). The final task to complete the discussion of the modern sector is thus to

define the total ex post production:

ZEt
Yt = Yi t di (3.19)
0

3.4.2 Traditional Sector

There exists a single firm in this sector, and there do not exist property rights over the
land.13 With total land endowment of the economy normalized to unity, these assump-
tions allow us to work with the following restricted form of the traditional technology

η
YT t = XT t HT t (3.12′ )

η
where XT t ≡ X̃T t .

Those who supply worker hours to this sector’s firm are assumed to be the owners

13. The alternative assumption of a unit mass continuum of identical firms does not alter the results under
perfect competition given the constant-returns-to-scale technology (3.12).

27
of the firm with equal ownership shares. Hence, these workers equally appropriate a
positive profit in addition to their income originating from labor supply. It turns out
η−1
that workers in the traditional sector earn, per unit hour, the average product XT t HT t .
In any equilibrium, then, we must have

η−1
XT t HT t = W t (3.20)

due to the perfect mobility of labor across sectors.

From (3.20) follows the mass NT t of workers employed in the traditional sector:

HT t
NT t ≡ (3.21)
ℓt

Note that a richer treatment with land rents would pose no serious difficulties given
the constant-returns-to-scale technology (3.12) and under, e.g., the law of primogeniture
that simplifies the intergenerational transmission of land ownership. In that case, those
who own, e.g., equal shares of land would appropriate the land rents in every generation,

leaving the land ownership to one of their surviving children.

3.5 Decision Problems

This subsection finally derives the decision problems solved by entrepreneurs and work-
ers. Three remarks follow before proceeding to the derivations:

First, notice that workers are indifferent between taking care of their children on
their own and hiring workers as entrepreneurs do so. In what follows, it is assumed that
workers take care of children with their own time.

Second, given preferences and the entire discussion above, the new elements needed
are the budget constraints individuals face.

Finally, note that only two problems are of interest, one for workers and one for
entrepreneurs. For workers, there is a unique problem because all earn W t per unit hour.

28
For entrepreneurs, the uniqueness of the problem follows from the fact that they all face
the same baseline productivity X t . That there exists only one problem solved by all
entrepreneurs in turn implies that entrepreneurs of generation t act symmetrically with
respect to certain choice variables. Then, once the uncertainty regarding the inventive

activity is resolved, a cross-section probability distribution characterizes the remaining

entrepreneur-level variables.

3.5.1 Workers’ Problem

Rewrite the utility function (3.2) for our representative worker who gives birth to bw t
children and consumes Cw t units:

Uw t ≡ Cw t + φ ln st bw t

Since this worker earns the unit wage of W t and spends ρbw t amount of worker hours

and ψbw t amount of goods for child care, her budget constraint reads

Cw t + ψbw t ≤ (ℓ t − ρbw t )W t

That Uw t is strictly increasing in Cw t leads this budget constraint to hold with strict
equality. Eliminating bw t via nw t = st bw t then implies

– ‚ Œ™ ‚ Œ
nw t nw t
Cw t = ℓ t − ρ Wt − ψ (3.22)
st st

Note that Cw t ≥ γ , following from (3.3), and (3.22) now determine the upper level of
nw t as in
(W t ℓ t − γ )st
– ™
nw t ∈ 1, (3.23)
ρW t + ψ

and workers, choosing nw t , seek to maximize

29
 – ‚ Œ™ ‚ Œ
nw t nw t
Uw t = ℓ t − ρ Wt − ψ + φ ln nw t (3.24)


st st

subject to (3.23).

3.5.2 Entrepreneurs’ Problem

The utility function (3.2) can be rewritten for entrepreneur i as

Ui t ≡ Ci t + φ ln st bi t

where Ci t and bi t respectively denote her consumption level and the number of children
she gives birth to. The budget constraint she faces reads

Ci t + ψbi t + ρbi t W t = Πi t (3.25)

where ψbi t + ρbi t W t indicates the total cost of having bi t children.

As in the derivation of the workers’ problem, Ui t is strictly increasing in Ci t , imply-

ing that the budget constraint above holds with strict equality. This allows us to rewrite

Ui t as
Ui t = Πi t − ψbi t − ρbi t W t + φ ln st bi t

The only worth-to-be-emphasized difference with the workers’ problem here is that
entrepreneur i’s decision towards inventive activity is taken under uncertainty, i.e. to

maximize EUi t = EΠi t − ψbi t − ρbi t W t + φ ln st bi t . Thus, the appropriate form of the

minimum consumption constraint for entrepreneurs’ problem is

ECi t ≥ γ

This inequality, the budget constraint above and EΠi t as in (3.17′′ ) imply that net
fertility ni t = st bi t is bounded above as in

30
    

 X t ‹Γ
 
ai t
exp Σai t Λ W ℓt − − γ st 
θξ (Kt )

 t 
ni t ∈  1,  (3.26)

 ρW t + ψ 


Therefore, entrepreneurs, choosing ni t and ai t , seek to maximize

 Γ !
Xt ai t
EUi t = exp Σai t Λ  ℓt −




Wt θξ K t
‚ Œ ‚ Œ
ni t ni t
−ψ −ρ W t + φ ln ni t (3.27)


st st

subject to (3.26) and

” —
ai t ∈ 0, a tmax (3.28)

where a tmax ≡ θξ K t ℓ t > 0 is the maximum possible arrival rate when entrepreneur

spends her entire labor endowment to inventive activity.

3.5.3 Occupational Choice

By solving the decision problems derived above, workers and entrepreneurs act opti-
mally. What completes the occupational choice with individual rationality is therefore
the following equal utilities restriction:

EUi t = Uw t (3.29)

An adult individual, taking all actions at the beginning of her adulthood, must be indif-
ferent between becoming an entrepreneur and becoming a worker.14

14. Note that, within this simple two-occupation framework with ex ante identical adult individuals, the
equilibrium mass of entrepreneurs is determined in the labor market residually; see below.

31
3.6 Market Clearing Conditions

The final task is to close the model through the market clearing conditions:
The market for worker hours clears via

ZEt
N t − E t − NT t ℓ t − N t − E t ρbw t = hw i t di + E t ρbi t (3.30)

where the R.H.S. and the L.H.S. of (3.30) respectively denote the total demand for and
the total supply of worker hours. The first term in the R.H.S. is the total amount of
hours employed in the modern sector firms, and the second term in the R.H.S. denotes

the amount of hours allocated to entrepreneurs’ child care. The supply of worker hours
is determined by the total available hours N t − E t − NT t ℓ t minus hours not supplied

to the market by workers due to child care.

The market for the consumption good clears via

ZEt
Ci t + ψbi t di + N t − E t Cw t + ψbw t




YT t + Yt =
0

where the R.H.S. denotes the total supply from traditional and modern sectors and the
L.H.S. denotes the total demand by entrepreneurs and workers.

32
 Chapter 4

Static, Dynamic, and Asymptotic Equilibria

This chapter defines and analyzes the equilibria of the model economy. The main pur-
pose in this chapter is to establish the analytical foundations of the model economy’s
equilibrium path from some initial period to the infinite future. Naturally, for things to
be interesting from a unified growth perspective, this equilibrium path should be long
enough to cover the transition from stagnation to growth in its entirety.

The static general equilibrium of the model is defined first. The important prop-
erties of this unique static general equilibrium are studied next. The dynamic general
equilibrium of the model is then defined as a sequence of static general equilibria. The
unique dynamic general equilibrium is further restricted to characterize the asymptotic

equilibrium of the model for t → ∞. A discussion on the asymptotic equilibrium and

its stability properties leads the chapter to a concluding section on the model economy’s
equilibrium path.

4.1 Static General Equilibrium

Definition 1: A static general equilibrium of this model, for any t ∈ N+ , is a collec-

tion {nw t , Cw t , HT t , NT t , YT t , E t , {ni t , ai t , h r i t , h mi t , zi t , Xi t , hw i t , Yi t , Πi t , Ci t }i ∈[0,Et ] , Yt }
of quantities and the wage W t such that, given the state vector (st , ℓ t , N t , K t , X t , XT t ),

• nw t solves the workers’ problem characterized by (3.23) and (3.24),

• ni t , ai t solves the entrepreneurs’ problem characterized by (3.26)-(3.28),

 • all adult individuals are indifferent between becoming an entrepreneur and becom-
ing a worker through (3.29),

• the market for worker hours clears through (3.30),15 and

• (3.5), (3.7), (3.8), (3.9), (3.15), (3.16), (3.18), (3.19), (3.12′ ), (3.20), (3.21), (3.22) and
(3.25) are satisfied. „

Proposition 1: There exists a unique static general equilibrium (SGE).

Proof — See Appendix A. „

4.1.1 Ex Ante Symmetry and Ex Post Heterogeneity

Since all entrepreneurs face the unique vector (W t , X t , K t , st , ℓ t ) of given variables in util-
ity maximization, they act symmetrically regarding the inventive activity:16

ai t = a t and hr i t = hr t for all i ∈ [0, E t ]

Notice that ex ante symmetry across entrepreneurs translates into ex post heterogene-
ity because of the stochastic nature of invention. Despite spending an equal amount of
research hours to inventive activity, entrepreneurs do not generate an equal number of
inventions. A fraction of them record no inventions, a fraction only one, and another
fraction two, and so on. Clearly, these fractions in cross-section establish a Poisson dis-

tribution under the assumption that E t is a large number. Specifically, for any arrival rate
a > 0, the ex ante probability of generating z inventions is equal to the ex post fraction
of entrepreneurs with z inventions.

15. Note that the market for the consumption good clears via Walras’ Law. Hence, the market clearing
condition for this market is not included in equilibrium defining equations.

16. That inventive effort is common across entrepreneurs in turn implies that they all spend an equal
amount h m t on management.

34
 The Poisson fractions naturally determine the variation of the operating productiv-
ity Xi t , of the demand for worker hours hw i t , of the volume of production Yi t , of the
level of profit Πi t and of the level of consumption Ci t across entrepreneurs. All these
distributions are identical to the unique (stationary) Poisson distribution Pois(a t ).

4.1.2 The Invention Threshold and the Industrial Revolution

The unique SGE of the model is characterized either by a t = 0 or by a t ∈ (0, a tmax ), and the
industrial revolution in the model is defined as the endogenously occurring switch from

the equilibrium regime of a t = 0 to that of a t > 0.

The threshold property of inventive activity directly follows from the solution to the
entrepreneurs’ problem through a rather straightforward algebra involving an applica-

tion of Kuhn-Tucker Theorem. Intuitively, the marginal cost of increasing the expected
number of inventions from zero to an infinitesimally small amount is a strictly positive
number that may well exceed its marginal return. This is due to the fact that entrepreneur
has to decrease her management input to increase her inventive effort. That is, the return

to and the cost of invention is not additively separable; see (3.17′′ ). Formally, we have the
following:

Proposition 2: The unique SGE of the model is characterized by an invention thresh-

old such that
 h  λ i−1
 0 if ξ K t ℓ t < θ σ 1−λ − 1


at =  λ −1 (4.1)
θξ K t ℓ t − σ − 1 otherwise


1−λ

Proof — See Appendix A. „

Notice that ℓ t , σ and λ increase the return to inventive activity and that θξ K t de-

creases the cost of it. Thus, these four determinants of inventive activity create threshold
effects accordingly.

35
 at

θξ (K t )ℓ t
 λ −1
θℓ t − σ 1−λ − 1

0 Kt
0 K ttr

Figure 4.1: The Arrival Rate as a Function of the Stock of Discoveries

For the rest of the analysis, it is useful to approach this threshold property by taking
K t as the endogenous state variable that determines whether a t = 0 or a t > 0. Suppose
that the inverse function ξ −1 (•) exists, and define the time-variant threshold originating

from Proposition 2 as
•  λ
−1 ˜
−1 −1
K ttr ≡ξ θ σ 1−λ −1 ℓ−1
t
(4.2)

Since ξ −1 (•) is strictly increasing, as implied by strictly increasing ξ (K) for all K < ∞,

K ttr is strictly decreasing in θ, σ, λ and ℓ t as expected. Figure 4.1 draws a t as a function

of K t where K ttr is as defined above.

From Proposition 2 follows the optimal allocation of entrepreneurs’ time to man-

agement. When invention is not optimal, entrepreneurs clearly spend their entire labor
endowment ℓ t to routine management to fully benefit from the profit opportunities un-
der constant productivity. Formally, we have


 ℓt if K t < K ttr
h mt = h
 λ i−1 (4.3)
θξ K t σ 1−λ − 1 otherwise


36
4.1.3 Productivity, Wage, and Output

The level of the real wage W t at the unique SGE, along with other things, determines the
level of economic development in this economy. It specifically determines the size of the
traditional sector and the optimal level of fertility. Thus, it is necessary to understand
how the real wage itself is tied to productivity.

The mapping from productivity to wage in the modern sector of the economy is

λ
W t = (1 − λ)1−λ λλ δ a t , K t , ℓ t X t (4.4)

where δ a t , K t , ℓ t is an auxiliary function defined as in

!1−λ
h  λ
 i at
δ a t , K t , ℓ t ≡ exp (1 − λ) σ − 1 at 1−


1−λ

θξ K t ℓ t

When inventive activity is not optimal (a t = 0), we have δ(0, •, •) = 1 implying that
λ
W t = (1 − λ)1−λ λλ X t . This simply corresponds to the unit price of a worker hour that
would prevail in a competitive model of occupational choice with Cobb-Douglas tech-
nology and without entrepreneurial invention.

When inventive activity is optimal (a t > 0), management input is tied to invention
technology through the optimal use of entrepreneurs’ time. W t in competitive equilib-
rium thus embeds this effect via δ(•, •, •) function. Notice, however, that the only source

of sustained growth in W t is still the sustained growth of X t ; a t and δ a t , K t , ℓ t are both

bounded above for any t .

Returning to the volume of output produced in the modern sector, the unique SGE
is characterized by

ZEt ‚ Œλ
λ λ
Yt ≡ Yi t di = E t X t ℓt δ at , Kt , ℓt (4.5)


1−λ
0

The total volume of output is proportional to the number of firms/entrepreneurs

37
simply because ex post heterogeneity obeys the well-behaved Poisson distribution.

4.1.4 The Structural Transformation

To see how the structural transformation is determined in this model, consider the labor
and the output shares of the modern sector defined as in

NT t Yt
ftN M ≡ 1 − and ftY M ≡
Nt YT t + Yt

In the unique SGE, the labor share NT t /N t of the traditional sector, measured in
individuals (not in worker hours), satisfies

1
1
‚ Œ‚ Œ
NT t XT t 1−η

= (4.6)
Nt ℓt Nt Wt

and the total volume of output produced in the traditional sector reads

XT1−η
t
YT t =
η (4.7)
W t 1−η

Recalling that the traditional and the modern sectors produce the same good, a higher
level of traditional sector productivity XT t implies a higher labor share NT t /N t of this

sector in contrast to the dual economy models in which (i) the land-based technology (or
agriculture) is used to produce food and (ii) there exists a minimum food consumption
restriction.

Population growth and longevity gains, on the other hand, decrease NT t /N t via the

push effect of the larger supply of hours ℓ t N t which is due to the dependency on fixed
land input.

The pull effect of the modern sector’s productivity is represented by W t . Figure 4.2
pictures the labor share of the traditional sector as a function of the real wage.

38
 NT t /N t

1
 € 1
XT t
Š 1−η
1
ℓt Nt W

0 Wt
W

Figure 4.2: The Labor Share of the Traditional Sector as a Function of Real Wage

4.1.5 Fertility and the Demographic Transition

The unique SGE of the model captures the phases of a demographic transition mainly
depending on the evolution of real wage W t .

Proposition 3: The unique SGE is characterized by identical fertility choice among all
adults, satisfying nw t = ni t = nt , such that


(W t ℓt −γ )s t φ+γ

 ρW t +ψ
if W t < ℓt
 h i
φ+γ φs t −ψ

φs t
nt = if W t ∈ , ρ (4.8)

 ρW t +ψ ℓt

φs t −ψ
1 if W t >

ρ

where the total births per adult, denoted by b t , is given by nt /st .

Proof — See Appendix A. „

(4.8) identifies the thresholds for fertility determined simply by the cost of and the
return to reproduction. For sufficiently low levels of potential lifetime income W t ℓ t ,
adult individuals’ priority remains as their own consumption. When this consumption

constraint binds, increasing W t and ℓ t allow them to increase their optimal net fertility
basically because they have more resources to spend on reproduction. Not surprisingly,

39
 nt

φℓt s t
ρ(φ+γ )+ψℓt

0 γ s t +ψ φ+γ φs t −ψ
Wt
0 ℓt s t −ρ ℓt ρ

Figure 4.3: Net Fertility as a Function of Real Wage

the parental desire for reproduction does not affect fertility during this first phase of the
demographic transition.

If the potential lifetime income continues to increase, it eventually becomes optimal

to decrease fertility since adults are now free to choose an optimal consumption higher
than the "subsistence" level of γ . The income effect vanishes completely at this second

stage of the demographic transition, and both net and gross fertility decline with increas-
ing W t . This phase of the demographic transition is thus characterized by the vanishing
role of adult longevity. The substitution effect, on the other hand, still implies a negative
effect of W t on fertility. Maintaining a level of net fertility that is greater than unity
becomes sufficiently costly for adults at the advanced stages of economic development,

i.e. W t > (φst − ψ)ρ. Thus, the third stage of the demographic transition records net
fertility equal to unity so that the level of adult population stabilizes. Net fertility as a
function of the real wage is shown in Figure 4.3.

4.1.6 The Supply of Entrepreneurship

The supply of entrepreneurship is central to the equilibrium path of the model economy
since the growth rate of K t is a function of the mass E t of entrepreneurs. This mass

40
satisfies ‚ Œ
NT t ρb t
E t = (1 − λ) 1 − − Nt (4.9)
Nt ℓt

Note that the mass of entrepreneurs satisfies (4.9) regardless of the invention threshold
and the thresholds of net fertility. There exist simply three determinants of the supply of
entrepreneurship: First, out of N t adult individuals, NT t individuals work in the prim-

itive sector. Second, a fraction of the total labor endowment is allocated to fertility per
adult. Finally, (1 − λ) is the ratio of entrepreneurs among the mass of individuals who
choose to work in the modern sector.

4.1.7 Output per worker and Output per capita

Two measures of the standard of living are output per worker and output per capita,
and it is useful to define them before proceeding to the analysis of the dynamic general

equilibrium.

Since every adult has b t children, the total population is equal to Pt ≡ (1 + b t )N t .

pw pc
Thus, output per worker yt and output per capita yt read

YT t + Yt YT t + Yt
ytp w = ytp c = (4.10)
Nt (1 + b t )N t

where YT t and Yt respectively satisfy (4.7) and (4.5).

4.2 Dynamic General Equilibrium

To define the dynamic general equilibrium, it is necessary to specify how the vector
(st , ℓ t , N t , K t , X t , XT t ) of state variables evolves from t to t + 1. For the exogenous state
variables, recall that the sequences {st , ℓ t }t ∈N+ are exogenously given. Next, adult popu-

lation N t and the stock of discoveries K t evolve respectively with (3.1) and (3.13). Thus,
the laws of motion to recover are those of X t and XT t .

41
 To derive the former, iterate (3.6) to obtain

Z Et
X t +1 = E t−1 Xi t di
0

Substituting Xi t with σ zi t X t and noting once again that the ex post fraction of en-
trepreneurs with z inventions is equal to the ex ante probability of generating z inven-

tions result in
a tz exp −a t

!
∞
X
X t +1 = X t σz
z=0 z!

This law of motion in turn reduces into the following after some arrangements as in
the proof of Lemma 1:
X t +1 = X t exp (σ − 1) a t (4.11)
 

Thus, as in Aghion and Howitt (1992) and others, the growth rate of (average) pro-
ductivity is explained by the size σ and the intensity a t of inventions.
η
Using (3.12) and the law of motion derived above, the growth rate of XT t = X̃T t can
be written as
XT t +1  η
= max ζ0 , exp ηζ1 (σ − 1) a t (4.12)
 
XT t

Definition 2: A dynamic general equilibrium of this model economy is a sequence

of static general equilibria, existing for all t ∈ N+ , together with the sequences
{N t , K t , X t , XT t }t ∈N++ , that satisfies the laws of motion (3.1), (3.13), (4.11) and (4.12) given

the sequences {st , ℓ t }t ∈N+ and the initial values (N0 , K0 , X 0 , XT 0 ). „

Proposition 4: There exists a unique dynamic general equilibrium (DGE).

Proof — See Appendix A. „

42
4.3 The Asymptotic Equilibrium

The asymptotic equilibrium of the model economy is the (unique) limiting SGE for
t → ∞. This limit, as argued earlier, corresponds to some distant future in historical

time. Motivated by growth empirics and by the "spirit" of the model, the asymptotic
equilibrium to be constructed and analyzed is the one with

• the positive growth of modern sector productivity and output,

• the constant level of population, and

• the declining traditional sector

Notice that, in this asymptotic equilibrium, output per worker and output per capita
exhibits exponential growth. One should however be careful about the nature of this

result: By definition, the model economy does not reach its asymptotic equilibrium in
finite time. The asymptotic equilibrium is conceptually and technically different from
a steady-state equilibrium if the latter is defined as a static equilibrium that is reached
in finite time and in which the bounded variables of the model remain constant from
some finite t to the infinite future. This sort of steadiness is satisfied by some variables of

the model in the unique DGE. Net fertility nt , for example, reaches its baseline level of
unity in finite time if the real wage grows to hit an endogenous threshold. Some other key
variables, however, do not reach their asymptotic values. The arrival rate a t , for example,
converges to
 λ −1
⋆
a ≡ lim a t = θ lim ξ (K t ) lim ℓ t − σ − 1
1−λ
t →∞ t →∞ t →∞
 λ

which reduces into a ⋆ = θ − σ 1−λ − 1 with ξ (K t ) → 1 and ℓ t → 1. The former limit
condition, as specified earlier, requires K t → ∞.

4.3.1 Two Preliminary Tasks

To construct the asymptotic equilibrium, two preliminary tasks have to be completed:

First, we need to preclude the uninteresting DGE where an industrial revolution is not

43
even possible. Next, we need to specify the condition(s) under which the traditional sector
does not operate when t → ∞.

The Possibility of an Industrial Revolution

The earlier restrictions that we put on ξ (K) function in (3.10) do not ensure K tt r < ∞;
see (4.2). In what follows, it is assumed that the inverse function ξ −1 (•) and its argument
 λ −1
θ−1 σ 1−λ − 1 ℓ−1
t
, for all t , are such that K tt r < ∞.

This assumption is less restrictive than it may sound because the simplest forms of
ξ (K) function that satisfy (3.10) also imply that K tt r < ∞ for sufficiently high θ, σ or λ
given any ℓ t .

The Decline of the Traditional Sector

For the DGE with an industrial revolution, what would ensure the decline of the tradi-
tional sector is the slower growth of XT t than that of W t for t → ∞:

‚ 1

1
Œ 1−η
X
 
Tt
f N M ⋆ ≡ lim ftN M = 1 − lim 
 
t →∞ ℓ⋆ N ⋆ t →∞ Wt


If the spillover from the modern to the traditional sector, i.e. from X t to XT t , never
η
becomes active in the unique DGE, then ζ0 < exp [λ(σ − 1)a ⋆ ] would be necessary as
λ
dictated by W t ∝ X t , (4.11) and (4.12). On the other hand, if the spillovers become
active at some t < ∞, what ensures the slower growth of XT t compared to that of W t for
t → ∞ is
λ
ζ1 < (4.13)
η
η
In what follows, it is assumed that (4.13) is satisfied. The former condition of ζ0 <
exp [λ(σ − 1)a ⋆ ], on the other hand, shall be of no importance since the value of ζ0 that

explains the long epochs of stagnation before the industrial revolution is always suffi-
ciently low to activate the sectoral spillover from X t to XT t ; see below.

44
 Table 4.1: Equilibrium Regimes

Regime Invention Fertility Spillover from X t to XT t

I at =0 nt incr. w/ W t inactive
II at =0 nt decr. w/ W t inactive
III at =0 nt = 1 inactive
IV at >0 nt incr. w/ W t inactive
V at >0 nt decr. w/ W t inactive
VI at >0 nt = 1 inactive
VII at =0 nt incr. w/ W t active
VIII at =0 nt decr. w/ W t active
IX at =0 nt = 1 active
X at >0 nt incr. w/ W t active
XI at >0 nt decr. w/ W t active
XII at >0 nt = 1 active

4.3.2 The Global Stability

Given the exogenous sequences {st , ℓ t }t ∈N+ , the model economy’s unique asymptotic
equilibrium characterized above is globally stable for the set of initial values and for the

set of model parameters that satisfy E t > 0 for all t ∈ N+ .

The difficulty here is that the dynamical system of the model is not simple enough to
allow us to rewrite it as an autonomous system of normalized variables. Instead, the anal-

ysis can only be carried out through a conditional dynamical system and with the help
of a phase diagram. What follows is a discussion of why the model’s asymptotic equi-
librium is globally stable, and the formal analysis of the conditional dynamical system is
presented in Appendix B due to its tediousness.

Recall that there exist two regimes of invention, three regimes of fertility, and two
regimes of the growth of the traditional sector productivity XT t . Hence there exist twelve

possible equilibrium regimes from which the model economy can start its evolution at
t = 0; see Table 4.1. The task is to understand why the economy eventually enters
Regime XII and stays in this regime for t → ∞.

The first key to understand why the economy remains in a regime of a t > 0 for large
enough t is the inevitability of an industrial revolution:

45
Proposition 5: Let the unique DGE feature a0 = 0. Then, there exists a period 0 <
t t r < ∞ such that a t t r −1 = 0 and a t t r > 0. That is, if the economy starts its evolution in
a period at which invention is not optimal, an industrial revolution inevitably starts at
some future period.

Proof — See Appendix A. „

Notice that the inevitability result of Proposition 5 implies that the arrival rate a t of

invention remains positive for t > t t r . That ℓ t is a non-decreasing sequence by assump-

tion and that K tt r is decreasing in ℓ t imply K t remains greater than K tt r for t > t t r . This
in turn implies that the arrival rate a t of inventions increases to a ⋆ after the industrial
revolution. Directly following is thus the positive asymptotic growth of X t and, thus,

of W t . Clearly, then, the asymptotic equilibrium does not exist in Regimes I, II, III, VII,
VIII, and IX.

The perpetual growth of W t is the main driver of the demographic transition. Re-
gardless of the stage of the demographic transition at t = 0, net fertility nt eventually
becomes equal to its baseline level of unity at some t < ∞, i.e. n ⋆ = 1. This in turn
implies that the level of adult population stabilizes at N ⋆ > 0. Hence, the asymptotic
equilibrium does not exist in Regimes IV, V, X, and XI as well.

Finally, the asymptotic growth of X t and W t eliminates Regime VI for t → ∞. As

stated earlier and explained below, ζ0 is sufficiently low such that, at some t < ∞, the

growth rate X t +1 /X t becomes sufficiently high to activate the sectoral spillovers; the
η
growth rate XT t +1 /XT t exceeds ζ0 .

4.3.3 The Asymptotic Rates of (Economic) Growth

Define the gross growth rate of the stock K t of useful discoveries using (3.13) and (4.9) as

h   i
NT t ρb t
K t +1 ωℓ t (1 − λ) 1 − N − ℓ N t
gK t ≡
t t
=1+ (4.14)
Kt Kt

With n ⋆ = s ⋆ = 1, gross fertility per adult attains the limit value b ⋆ = 1 in finite time.

46
 K t +1

K t + ωℓ t E t
45◦
K t + ωℓ⋆ E ⋆

0 Kt
0

Figure 4.4: The Process of Collective Discovery for t < ∞ and for t → ∞

Since NT t converges to zero, gK t satisfies

ω(1 − λ)(1 − ρ)N ⋆

gK⋆ ≡ lim gK t = 1 +
t →∞ lim t →∞ K t

where (E ⋆ ≡)(1 − λ)(1 − ρ)N ⋆ is the asymptotic value of the mass E t of entrepreneurs.
Clearly, E ⋆ > 0 implies both gK t > 1 for t < ∞ and gK⋆ = 1 for t → ∞. These dynamics

are pictured in Figure 4.4.

Let gX t ≡ X t +1 /X t denote the gross growth rate of the modern sector productivity.

Using the results derived earlier, we can write this growth rate as

•   λ −1 ‹˜
gX t = exp (σ − 1) θξ (K t )ℓ t − σ 1−λ − 1

In the limit, then, the asymptotic growth rate gX⋆ is a function of technological parameters
θ, σ, and λ, all having definite interpretations.
pw pc
Returning to output per worker (yt ) and output per capita (yt ), first note that the

47
asymptotic growth of these variables is explained entirely by the growth of X t : For t →
∞, the traditional sector declines, and adult population and fertility are fixed. Recalling
λ
that the modern sector output is proportional to X t , we can conclude that the asymptotic

growth rate gy⋆ of living standards is equal to

•   λ −1 ‹˜ € Šλ
gy⋆ = exp λ(σ − 1) θ − σ 1−λ − 1 = gX⋆

In the asymptotic equilibrium, then, output per capita and output per worker grow at
a rate proportional to but lower than that of labor productivity where the stock of useful
discoveries and the level of population do not exhibit growth. Formally, we have

gX⋆ > gy⋆ > 1 gK⋆ = gN⋆ = 1

4.4 The Absence of the Weak and the Strong Scale Effects

An important feature of the model economy is to be noted before proceeding to the

construction of the equilibrium path: Not surprisingly, the weak and the strong scale
effects of population size are absent in this economy. In fact, the absence of scale effects

on output per worker and output per capita holds not only at the asymptotic equilibrium
but also along the transition.

To see why there does not exist a weak scale effect, recall that (i) the volume YT t of
the traditional sector output is not related with adult population N t , and (ii) the volume

Yt of the modern sector output is proportional to E t (which itself is proportional to N t ).

That is, if the level of population is higher, there exists a larger mass of entrepreneurs,
ceteris paribus, and, in the absence of any direct positive externality of E t or N t on total
pc pw
output, the SGE levels of yt and yt are not affected positively by N t .

The strong scale effect, on the other hand, is absent because of the following:

Even though the model does not explicitly incorporate the horizontal dimension of
technological progress (or product innovation) as in the second-generation Schumpete-

48
rian models of Young (1998), Peretto (1998a), Aghion and Howitt (1998, Ch. 12) and
Dinopoulos and Thompson (1998), that the innovative sector of the economy is inhab-
ited by a mass E t of independently innovating firms still implies that the total research
effort of the economy is spread thinly across modern sector firms. Given that there does

not exist a direct link from population level to productivity growth, the transition and
pc pw
the asymptotic growth rates of yt and yt do not change with the level of population.17

4.5 The Equilibrium Path: From Stagnation to Growth

We can now construct an equilibrium path of the model economy from some distant past
t = 0 to some distant future t → ∞. Specifically, the interest here is on an equilibrium
path with an initial SGE that exists in Regime I. This is clearly the most interesting equi-
librium path from a unified growth perspective. The stock of useful discoveries is small

enough to make invention not optimal, productivity increases only in the traditional sec-
tor and at a constant minuscule rate, and the level of the modern sector productivity is
low enough to make fertility an increasing function of the real wage.
There surely exists a set of initial values of state variables that further characterize the
initial SGE in Regime I as follows:

• net fertility and the rate of population growth are low,

• gross fertility is high,

• the traditional sector’s labor and output shares are large, and

• the levels of the standards of living are low.

The model inputs, i.e. structural parameters, initial values, and exogenous variables,
must be such that this initial equilibrium is sustained from t = 0 to some future period
under quasi-statis.

17. Strictly speaking, there exists a type of strong scale effect on the growth of K t . This is discussed below
in Section 7.5.

49
 First, realistically suppose that st and ℓ t are constant respectively at their historically
lowest levels s0 and ℓ0 for a long episode of history until the industrial revolution is near.
This returns a level of net fertility which is equal to

(W0 ℓ0 − γ )s0
n0 =
ρW0 + ψ

Notice that, for sufficiently low W0 , net fertility and the rate of population growth are

sufficiently low. Next, recalling that the labor share of the traditional sector is equal to

1
1
‚ Œ‚ Œ
NT t XT t 1−η

=
Nt ℓ0 N t W0

a restriction needs to be put on the growth of XT t to imply a quasi-static level of NT t /N t .

It turns out that if ζ0 satisfies
η

ζ01−η = n0

the labor share NT t /N t remains stable at its historically high level for long periods; slowly
increasing population requires slowly increasing traditional sector productivity at this

equilibrium.
pw
The model thus explains the stagnation of output per worker yt and output per
pc pw pc
capita yt at their historically lowest (constant) levels y0 and output per capita y0 if ζ0
satisfies the above condition, but the critical level of ζ0 is not arbitrary: If there is a stable

quasi-static equilibrium of sectoral shares of labor and output, this equilibrium has to be
characterized by stagnation of living standards.18

An important remark is in order: The initial quasi-static stagnation equilibrium just

characterized is not strongly Malthusian since the real wage that determines the level of

optimal fertility changes with the modern sector’s productivity but not with the level of

18. Regarding the knife-edge type of restriction that is put on ζ0 , note that endogenizing the productivity
growth of the traditional technology via learning-by-doing as in Strulik and Weisdorf (2008) would
imply an automatically stabilizing dynamical system at the stagnation equilibrium just characterized.

50
population. Increasing population in this model does not decrease the real wage. This
strong Malthusian link would be constructed by allowing the modern and the traditional
sector wages to differ in a richer framework of rural-urban migration.
Returning now to the equilibrium path, stagnation in strict sense ends when the econ-

omy eventually hits its invention threshold. Collective discovery, operating all along
during the stagnation, eventually closes the knowledge gap. The real wage starts increas-
ing, leading to increasing gross and net fertility. Increasing inventive effort initiates the
sectoral spillover from the modern to the traditional sector at some period, but faster
population growth and technological progress in the modern sector still imply declining

output and labor shares of the traditional sector. Once the modern sector productivity
is sufficiently large, gross fertility starts declining because adults’ time is now sufficiently
expensive. Decreasing population growth and still increasing inventive effort leads even
faster growth of output per capita/worker. On the other hand, urbanization and industri-

alization continue. At some advanced stage of economic development with a sufficiently

high modern sector productivity, net fertility becomes equal to its baseline level of unity
because the adult individual now finds it optimal to sustain reproductive success at its
limit. The stabilizing population in turn implies that the growth rate of the stock of

useful discoveries slows down. Even though the existing entrepreneurs keep discovering
new knowledge, each generation’s contribution gets marginally smaller compared to the
existing stock of discoveries. Once this slowing down starts, the increase of the inventive
effort starts decelerating. With constant population and constant fertility, the growth of
modern sector productivity converges to its maximum. In the future, according to this

model without material resource constraints other than that of labor, humanity faces no
limits in increasing prosperity.

51
 Chapter 5

The Timing of the Industrial Revolution: Some Tentative Results

The consensus view of world economic history emphasizes the lateness of modern eco-
nomic growth: The Industrial Revolution in England started around 50000 years later
than the rise of modern human populations, i.e. populations that share cultural univer-
sals such as language, art, religion, and toolmaking. If one rescales the history of mod-

ern human populations to 365 days, the Industrial Revolution would have had occurred
around 44 hours ago.

The model constructed and analyzed above suggests that living standards may stag-

nate for several millennia if purposeful invention in the modern sector is too costly, i.e. if
lives remain sufficiently short and the stock of useful discoveries remains sufficiently
small.

This chapter presents some tentative analytical results on the timing of the industrial
revolution in the model. The timing question basically asks which factors are most essen-
tial in explaining the length of time passed with stagnating standards of living in a given
economy for some fixed initial period.

From a methodological point of view, some may find the timing question very am-
bitious. This is due, in part, to the simplicity of the models that assume away several a
priori important aspects of the transition from stagnation to growth as the present model

does and, in part, to the complexity of the phenomenon being dealt with. Economic
historian Jones (2010, p. 245), e.g., claims that
 There is no determinate solution to the puzzle of why the industrial revolution took
place, and where and when it did so. All that can be achieved is a narrowing of the
range of possible mixes.

The model, by providing answers to why the industrial revolution took place and
why that late, narrows the range of possible mixes in Jones’ (2010) terms. Yet, answering
questions such as "Why England, but not France or China?" and "Why 18th century, but
not the 14th?" is at best harder.

First of all, the timing question may not be simply separable from the location ques-
tion because a mechanism that explains the timing of the industrial revolution for a given

economy may provide insights to understand the absence of an industrial revolution in

another economy. As in Mokyr (2002), indeed, the collective discovery by entrepreneurs
not only explains the lateness of the industrial revolution in England but also implies that
China, with its infamously large bureaucracy and very low urbanization, did not benefit

from collective learning because of a lack of entrepreneurship.

Second, as suggested long ago by Crafts (1977), luck may have played a key role

to put England in front of France and other Western European nations for a few
generations in industrialization. The model-based cross-sectional calibration results of
Voigtländer and Voth (2006) indeed imply that France had a much higher likelihood of
industrialization than China on the eve of Industrial Revolution. A valid argument thus
can be made to reformulate the location question as "Why Western Europe, but not

China?"

Yet, there exists another difficulty about how pre-industrial economies should be
compared with regard to the timing of the industrial revolution. Pomeranz (2000), e.g.,
argues that, due to the regional diversity of big geographical areas such as Western Eu-
rope and China, the comparisons should be made between economies of appropriate
size. England, e.g., should be compared not with China as a whole but its most econom-

ically advanced Lower Yangzi region. In light of these difficulties, the tentative analytical
results on the timing of the industrial revolution should better be read as suggestive, not

53
conclusive.

5.1 Preliminaries

Since we direct our attention to the DGE with a0 = 0, it is useful for future reference to
note that the modern sector productivity X t and the real wage W t stagnate respectively
at
λ
X0 > 0 W0 = (1 − λ)1−λλλ X 0 > 0

Clearly, W0 is increasing in λ, for any X 0 > 0, if λ ∈ (0.5, 1).

5.1.1 First-Order and Second-Order Effects

One difficulty associated with the timing question is that the period t t r at which the
industrial revolution starts does not have a closed-form solution in terms of the model’s
parameters and state variables. Thus, the feasible analysis here is to question whether

a higher value of a relevant model parameter or of an exogenous state variable, ceteris

paribus, delays or hastens the start of the industrial revolution from the perspective of some
given period. That is, we fix a period t < t t r and ask, for each element of the vector

(st , ℓ t , γ , ψ, φ, ρ, η, θ, σ, λ, ω),

whether the industrial revolution would be delayed or hastened if the chosen element
of the vector is higher than some benchmark value given the endogenous state variables.

These are labeled as first-order effects.

The second-order effects, remaining analytically implicit, are the ones that run through
the endogenous state variables. That is, when a model parameter is higher, it affects not
only the SGE of any fixed period t < t t r but also the SGE of t +1, t +2, and so on. As we

shall see in the next chapter, these second-order effects are indeed substantial especially
through the population growth.

54
5.1.2 Threshold vs. Growth Effects

There are two types of first-order effects to be distinguished regarding the timing ques-
tion: First, there exists an endogenous and time-varying threshold K tt r that defines the

knowledge gap
•  λ
−1 ˜
−1 −1
K t − K tt r = Kt − ξ θ σ 1−λ −1 ℓ−1
t

for any given K t . Clearly, for the inevitable industrial revolution to start, this gap should

be greater than or equal to zero at t t r . Hence, the threshold effect is related with the
question of how far away the industrial revolution is for some t < t t r .

The growth effect, on the other hand, is related with the question of how fast the
economy moves towards its invention threshold to decrease its knowledge gap, and the

(gross) growth rate gK t of K t determines this speed for any given K tt r . Rewrite this growth
rate by substituting NT t /N t and then W0 as

 1 
Œ  1−η
1
‚ ‚ Œ
 XT t ρb t  N t
gK t 1 − ℓ N 
= 1 + ωℓ t (1 − λ)   − (5.1)
  
1−λ λ λ ℓ  K
t t (1 − λ) λ X t
0
t

Notice that gK t changes with the level of optimal gross fertility b t . This separates the
analysis further into two cases, i.e. the start of the industrial revolution (i) in the regime
where fertility is increasing with the real wage and (ii) in the regime where fertility is
decreasing with the real wage.19 Therefore, we should rewrite gK t explicitly for each case.

For the regime where fertility is increasing with the real wage, (5.1) becomes

 1
‚ Œ
1
‚ Œ‚ Œ 1−η
γ ρ + ψℓ t X Tt  Nt
gK t = 1 + ω(1 − λ)  − (5.1′ )

ρW0 + ψ Nt W0

Kt

19. The regimes in which n t = 1 are of minor importance for the present analysis since, for the economy
to be in such a regime, the real wage must already be sufficiently high.

55
 Table 5.1: The Timing of the Industrial Revolution: Case 1 (nt incr. w/ W t )

Experiment Threshold Effect Growth Effect

Higher st No Effect No Effect
Higher ℓ t Hastening Hastening
Higher γ No Effect Hastening
Higher ψ No Effect Hastening
Higher φ No Effect No Effect
Higher ρ No Effect Delaying
Higher η No Effect Hastening
Higher θ Hastening No Effect
Higher σ Hastening No Effect
Higher λ Hastening Ambiguous
Higher ω No Effect Hastening

and, for the regime where fertility is decreasing with the real wage, we have

 1
‚ Œ
 ℓ t (ρW0 + ψ) − ρφ 1
‚ Œ‚ Œ 1−η
XT t  Nt
gK t = 1 + ω(1 − λ)  − (5.1′′ )
ρW0 + ψ Nt W0

Kt

5.2 Results

The partial derivatives of K tt r and gK t with respect to model parameters and exogenous
variables identify the threshold and the growth effects given the set (N t , K t , X t , XT t ) of
endogenous state variables.

Tables 5.1 and 5.2 summarize the results of the analysis. "No Effect" entries indicate
a partial derivative of zero. An "Ambiguous" entry on the other hand indicates a partial

derivative with an ambiguous sign.

In Tables 5.1 and 5.2, the threshold effects are identical and simply follow from strictly
increasing ξ −1 (•). As discussed earlier, ℓ t , σ and λ increase the return to inventive activ-

ity, and θ decreases the cost of it. The remaining model parameters and st do not generate
threshold effects.

Returning to the growth effects, first note that st does not generate growth effects,
neither in Case 1 nor in Case 2. This follows from the fact that gross fertility does not

56
 Table 5.2: The Timing of the Industrial Revolution: Case 2 (nt decr. w/ W t )

Experiment Threshold Effect Growth Effect

Higher st No Effect No Effect
Higher ℓ t Hastening Hastening
Higher γ No Effect No Effect
Higher ψ No Effect Hastening
Higher φ No Effect Delaying
Higher ρ No Effect Delaying
Higher η No Effect Hastening
Higher θ Hastening No Effect
Higher σ Hastening No Effect
Higher λ Hastening Ambiguous
Higher ω No Effect Hastening

change with the survival probability in this model.20

A higher value of adult longevity ℓ t has the hastening growth effect in both cases.

Entrepreneurs generate a larger number of useful discoveries if they live longer.

A higher value of ω, implying a higher "quality" of the process of collective discovery,

also hastens the industrial revolution in both Case 1 and Case 2.

A larger labor exponent η of the traditional technology implies a lower share of the
traditional sector and has the hastening growth effect in both cases.

More interesting are the effects of the parameters that determine gross fertility. The

unit good cost of reproduction ψ decreases gross fertility in both cases. Accordingly, a
higher value of ψ has hastening effects in both cases because a smaller volume of labor
endowment is allocated to child rearing due to the decreasing level of fertility. The unit
time cost parameter ρ, however, have the reverse effect. Even if a higher value of ρ
decreases gross fertility by increasing the cost of reproduction, it also directly increases

the total amount of hours allocated to child rearing. ρ generates a delaying growth effect
in both cases since the direct effect dominates.

A higher value of γ implies a lower level of gross fertility by tightening the budget

constraint of adults in the regime where fertility is increasing with the real wage. This

20. Notice that s t does have a second-order (hastening) growth effect through N t .

57
is a hastening growth effect, but γ does not generate a growth effect in Case 2. Instead,
a growth effect in Case 2 is generated by φ. If adults prefer to have more children, this
translates into a delaying effect, again, by increasing the total amount of hours allocated
to child rearing.

The set of parameters that do not generate growth effects include θ and σ. Both are
technological parameters governing the return to the inventive effort, but they do not
affect the growth rate of K t in any regime before the industrial revolution.
The labor exponent λ of the modern sector technology, in addition to its hastening
threshold effect, creates a growth effect. The direction of this growth effect however

remains ambiguous. In both Case 1 and Case 2, a higher value of λ may imply either a
lower or a higher mass of entrepreneurs because, technically, both fertility and the labor
share of the traditional sector change with W0 and hence with λ. For any given level
of W0 , a higher value of λ would imply a lower mass of entrepreneurs since the level of

optimal demand for worker hours by entrepreneurs is larger on average. Yet, since W0
changes with λ, we have
∂ gK t
⋚0
∂λ

in both Case 1 and Case 2.

58
 Chapter 6

The Quantitative Analysis

This chapter presents a quantitative analysis of the model economy. The purpose of this
analysis is threefold: First of all, a quantitative analysis of the model, disciplined empiri-

cally via formal calibration, is essential to evaluate the model’s success in explaining the
phases of economic development from pre-industrial times to the contemporary era. Sec-
ond, since the very long-run evolution of the model economy is of interest, this chapter
presents the simulations of the calibrated model economy and studies the transition to

its previously described asymptotic equilibrium. Finally, the timing question is revis-
ited quantitatively, and some counter-factual experiments, again based on the calibrated
model, are implemented to conclude which factors are most important in determining
the timing of the industrial revolution. These experiments are necessary since the tenta-

tive results presented in the last chapter do not reveal the overall timing effects of model
inputs.

6.1 Preliminaries

The empirical counterpart of the model economy is that of England since it is the first
industrialized economy in the world. Fortunately, England is also the economy of which
the largest data set on certain variables of interest is available compared to the other
Western European economies.
 20 100
Real Net National Income p. c. (right axis)

15 80

1860 = 100
millions

10 60

5 40
Population (left axis)

0 20
1200 1300 1400 1500 1600 1700 1800 1900

20 100
Real Wage (right axis)

15 80

1860 = 100
millions

10 60

5 40
Population (left axis)

0 20
1200 1300 1400 1500 1600 1700 1800 1900

Figure 6.1: Real Income, Real Wage, and Population in England: 1200-1860

Data Source: Clark (2009).

6.1.1 Time, Generational Growth Rates, and the Initial Period

Throughout the analysis, the model’s initial period is denoted by t = 1 and the length
of a period is taken to be 25 years, i.e. the lifetime of one generation. The growth rates
reported are hence generational growth rates per 25 years. The conversion to the annual

rates is not necessary since the interest is limited with the long-run evolution of the econ-
omy.

60
 Real N.N.I. per capita Population
700 50

600
40
500

30
400

millions
index

300
20

200
10
100

0 0
1600 1700 1800 1900 2000 1600 1700 1800 1900 2000

Figure 6.2: Real NNI per capita and Population: 1650-2000

Notes: The original data source for both series is Clark (2009). Since the raw data
are decennial, mid-decade levels for 1675, 1725, and so on are calculated using simple
arithmetic average between closest decennial data. That is, the level of real N.N.I. per
capita in 1675, e.g., is equal to the average of those in 1670 and 1680 and so on.

The initial period t = 1 of the model economy is matched with the year 1650. The

reasoning behind this choice is threefold:

First, as argued earlier, the model is not designed to capture the Malthusian cycle be-
tween the real wage and the level of population. Since this type of Malthusian dynamism

in England ends at around the year 1650 after which both population and living standards
exhibit exponential growth, matching t = 1 with the year 1650 is not restrictive for the
present purposes (see Figure 6.1).

Second, the year 1650 is early enough to let the model economy to reveal the dy-
namics before the industrial revolution. Normalizing the start date of the latter to 1750,
matching t = 1 with the year 1650 gives us four generations, i.e. a century, to analyze the
growth of K t and N t before the industrial revolution.

Finally, satisfactorily rich data exist for England only for the period after mid-1500s.
As we shall see next, this is true both for the demographic variables and for the labor and
the output shares of the traditional sector.

61
 Traditional Sec. Labor Share Traditional Sec. Output Share
70 70

60 60

50 50

40 40
percent

percent
30 30

20 20

10 10

0 0
1600 1700 1800 1900 2000 1600 1700 1800 1900 2000

Figure 6.3: The Decline of the Traditional Sector: 1650-1975

Notes: The original data for the labor share of the modern sector is collected by
Clark (2001) for England, covering the period 1565-1865, and by Maddison (1995)
for the United Kingdom, covering the period 1820-1992. The data on the output
share of the modern sector is generated by Bar and Leukhina (2010a) using the raw
data from Clark (2001), Clark (2002), and Mitchell (1975). See Appendix A.1 of
Bar and Leukhina (2010a) for details.

6.1.2 The Data

Living Standards and Population

The quantitative analysis of the model, first of all, requires a measure of living standards.

This measure is real net national income per capita (real N.N.I. p.c.) estimated by Clark
(2009).21 The raw decennial data is next used to construct the time series from 1650 to
2000 with 25 year intervals by taking arithmetic averages for 25th and 75th years of any
century.

The population in England is another key macroeconomic aggregate, and the raw

data series is again borrowed from Clark (2009). This time series decennially runs from
1200 to 1860. To complete this series, the United Kingdom’s (decennial) census data for

21. Specifically, nominal net national income is deflated with the price index of net domestic output for
the period 1200-1860, and the index of real income per person is linked with the earlier series for 1860
and beyond. See Tables 28 and 34 in Clark (2009) for more details.

62
England from 1871 to 2001 are used.22 Then, the decennial series is utilized to construct
the dataset of 1650-2000 period with 25 year intervals using the same principles applied
to the income data.

The Decline of the Traditional Sector

The measures of the structural transformation are the labor and the output shares of
the traditional sector. The declines of these ratios are associated respectively with ur-
banization and industrialization processes as in Bar and Leukhina’s (2010a) study, and

Bar and Leukhina’s (2010a) dataset on the labor and the output shares of the modern
sector is used to generate the associated shares.23 Figure 6.3 pictures the decline of the
traditional sector for the period 1650-1975.

Survival Probability and Adult Longevity

Figure 6.4 pictures the actual data and the fitted values for survival probability st and
(normalized) adult longevity ℓ t . The fitted values for the period 1650-2250 are needed for
the long-run simulations and used instead of the actual data to design counter-factual ex-

periments in a systematic manner. Before discussing how these fitted values are obtained,
however, it is necessary to clarify the sources and the generation of the actual data.

The actual data series for the survival probability for ages 0-25 is based on the data
used by Bar and Leukhina (2010a).24 Bar and Leukhina’s (2010a) finalized data for st runs
from the mid-year 1612.5 to the mid-year 1987.5 in 25 year intervals. Setting the survival
probability for the mid-year 2012.5 to 0.995, the data series for the period 1625-2000 is

22. Strictly speaking, the census year is matched with the decade it starts; population of the year 1870 in
the completed series is the level of population recorded in 1871 census, and so on. Population data
from the U.K. censuses are available online at www.statistics.gov.uk/ website.

23. The dataset used by Bar and Leukhina (2010a) is obtained via personal communication with the au-
thors.

24. The original data sources are Wrigley et al. (1997) and Human Mortality Database. The latter is acces-
sible online at www.mortality.org/ website.

63
 Table 6.1: The Regression Results for the Exogenous Variables
The Logistic NLS Fit of s t The Logistic NLS Fit of ℓ t

Estimate 95% Con. Int. Estimate 95% Con. Int.

µs 0 0.63 [0.62, 0.65] µℓ0 0.42 (0.40, 0.44)
µs 1 1.15 [0.82, 1.48] µℓ1 0.37 (0.31, 0.44)
µs 2 12.83 [12.54, 13.11] µℓ2 15.75 (15.38, 16.13)

S.S.E. 0.00502 S.S.E. 0.00432

R.M.S.E. 0.01967 R.M.S.E. 0.01823
Adj. R2 0.97820 Adj. R2 0.98230

generated using arithmetic averages.25

The data on adult longevity is constructed as follows: Using the period life expectancy
data of Wrigley et al. (1997) for the periods before 1825 and of the United Kingdom Na-
tional Statistics for the periods after 1850, the period life expectancy at age 25, denoted
by e25 , is retrieved. Then, setting the maximum lifespan of humans to 100 years, the nor-

malized adult longevity corresponding to ℓ t in the model, is generated by dividing e25 to

75 for all periods.

The fitted values for st and ℓ t are obtained by estimating the following simple logistic
equations via Nonlinear Least Squares

1 − µs 0
st = µs 0 +  + εs t
1 + exp −µ s 1 (t − µ s 2 )


1 − µℓ0
ℓ t = µℓ0 + + εℓt
1 + exp [−µℓ1 (t − µℓ2 )]

where t takes integer values starting from unity and εs t and εℓt are error terms. Table 6.1
summarizes the estimation results.

25. We shall see below that the value of s t in 1625, i.e. in the model period t = 0, is needed to obtain a
model-based data counterpart for fertility.

64
 Survival Probability
1

0.9

0.8

0.7 Logistic Fit

Data

1600 1700 1800 1900 2000 2100 2200 2300

Adult Longevity (Normalized)

1

0.8

0.6

0.4
1600 1700 1800 1900 2000 2100 2200 2300

Figure 6.4: Survival Probability and Adult Longevity: Data vs. Fitted Values

Notes: See the text for a description of actual data. Fitted values are obtained through
Nonlinear Least Squares estimation of the logistic functions introduced. See Table
6.1 for the regression results.

Fertility Data

A data series needed for the quantitative analysis is that of gross fertility. The first alterna-

tive here would be to use the actual data on total fertility rate since it is explicitly defined
as a model variable. This however is not a completely fair way to judge the model’s ca-
pability of explaining fertility and population dynamics since the demographic structure
of the model is extremely simple with two generations, asexual reproduction, and a com-
mon gross fertility level for all adult individuals. Another alternative would be to derive

the model-based counterpart of the crude birth rate or the general fertility rate while
making some necessary corrections using the survival probabilities for different ages and

65
 200
TFR
4.5
Rep. Lev.

4 40

3.5

3
100
20
2.5

CBR (left axis)

2 GFR (right axis)

1650 1700 1750 1800 1850 1900 1950 2000 1650 1700 1750 1800 1850 1900 1950 2000

Figure 6.5: The Measures of Gross Fertility in England: 1650-2000

Notes: Crude birth rate (CBR) and general fertility rate (GFR) measure the annual
number of live births respectively per 1000 people and per 1000 women of childbear-
ing age. The dashed line indicates the replacement level for a survival probability of
unity.

mid-year population levels. Yet, there is a third alternative which is clearly less tedious
than the second one. Since we have actual data series of population and survival prob-
ability, the model-based counterpart of gross fertility per adult can be easily obtained

using
1 1 + b t −1
‚ Œ‚ Œ‚ Œ
Pt
bt = −1
Pt −1 st −1 b t −1

in a recursive way where this equation follows from Pt = (1 + b t )N t and N t +1 = st b t N t .

The model-based data counterpart of total fertility rate is thus 2b t .

Figure 6.5 contrasts model-based total fertility rate described above with the actual
data on general fertility and crude birth rates collected by Bar and Leukhina (2010a).26
Clearly, and not surprisingly, the derived data, pictured in the upper-panel of Figure 6.5,

closely represents the dynamics of gross fertility measured by general fertility and crude

26. To generate the model-based total fertility rate data, smoothed population series and the logistic fit of
survival probability is used. 2b t −1 is taken to be 3.50. This is the maximum level of total fertility rate
that implies a non-decreasing adult population after 1650. Moreover, the dependency of b t series on
this initial value dies out rather quickly after 2 to 3 periods.

66
birth rates. More importantly, if the model economy is successful in explaining the dy-
namics of model-based total fertility rate, it comes closer to explain the dynamics of the
level of population since st is exogenous.

The Share of Entrepreneurs

The last piece of data to be introduced, actually a single data point, is of the share of
entrepreneurs in adult population. The ratio 1 − e t = (N t − E t )/N t in the model is
the share of all working-age individuals employed as (manual) workers. Thus, the share

of individuals working as non-manual workers such as managers and business owners

corresponds to the ratio e t in the model. Fortunately, the data collected by Routh (1987)
exactly measures this: In 1975, 26.075% of all working-age individuals in England were
non-manual workers.

6.1.3 Parameterizing the ξ (K) Function

The last task before proceeding to the discussion of how the model parameters and initial
values are calibrated is to specify an explicit form for ξ (K) function. Since there does not

exist an empirically guided way of characterizing ξ (K) function beyond the restrictions
put earlier via (3.10), the quantitative analysis is based on the simplest functional form
satisfying these restrictions:
1
ξ (K t ) = 1 −
1 + Kt

The reason behind specifying ξ (K t ) without introducing a new structural parameter

is an identification difficulty. θ > 0 that represents the extrinsic productivity of unit
inventive effort is already separated from ξ (K t ), and a parameter changing ξ (K t ) for a
given value of K t would not be identified without some additional data on discovery-
invention relationship.27

27. Note that another very simple form of ξ (K t ) that satisfies (3.10) is ξ (K t ) = 1 − exp(−K t ). Using this
functional form however does not change the qualitative nature of results.

67
6.1.4 Calibration

To simulate the model via forward recursion, all model parameters and the initial values
of all endogenous state variables for t = 1, i.e. (N1 , K1 , X 1 , XT 1 ), must be calibrated to

unique values. The purpose of the calibration exercise is to let the actual data determine
the calibrated values wherever possible.

Two difficulties with the calibration are the following: First, the model’s asymptotic

equilibrium, as an equilibrium at which most variables are fixed at their asymptotic levels,
cannot be used for the calibration basically because actual data do not exist for t → ∞.28
This difficulty implies that a simulated method of moments type of calibration strategy
must be used; an algorithm that solves the model for a given set of parameters and initial

values and minimizes a quadratic form of deviations between model predictions and ac-
tual data. Second, since there exist a total of 12 equilibrium regimes and the purpose is to
match the observed timing of the industrial revolution and the demographic transition,
there exist hidden constraints on parameters and initial values. That is, the benchmark set

of parameters and initial values must ensure that the economy is in Regime I at t = 1 and
an industrial revolution starts at t = t t r = 5 that corresponds to 1750.

The calibration strategy is mixed: First, the value of a parameter is borrowed from

the literature to ease the calibration exercise. Second, each initial value is normalized or
calibrated from the data. Third, some identified parameters are solved using the relevant
data points that define a set of data targets. Finally, the remaining parameters are jointly
calibrated. That is, extending the set of data targets, a numerical algorithm is used to
determine the values of these parameters that minimize an objective function measuring

a distance between model predictions and data targets.

The labor exponent η ∈ (0, 1) of the traditional technology is set to 0.537 which is cal-
ibrated by Bar and Leukhina (2010a) for England; the traditional technology postulated

28. By featuring n t = 1 and f tN M = 0, the model’s asymptotic equilibrium is in fact not completely in-
formative about all of the parameter values. Strictly speaking, though, the asymptotic growth rate of
modern sector productivity is restricted to identify one of the parameters; see below.

68
by Bar and Leukhina (2010a) isolates the input of labor hours just as in this model.

The initial value N1 of adult population, measured in millions, is solved from N1 =

P1 /(1+b1 ) where both population P1 and gross fertility per adult b1 are actual data values.
The calibrated value of N1 is 2.0718 millions.

K1 is normalized to unity. Notice that K1 and ω both affect the timing of the indus-
trial revolution. The normalization thus identifies ω given (i) the period at which the
industrial revolution starts, and (ii) other determinants of the growth of K t .

Also normalized to unity is X 1 . The arbitrariness here does not matter quantitatively
since (i) output (per capita) data is available as an indexed variable and (ii) there always
exists a value of γ > 0 that implies a given level of initial fertility for any value of X 1 .

When the model predictions and the actual data are contrasted below, the model predic-
tions of output (per capita) and real wage, obtained under the normalization X 1 = 1, are
reindexed without any affect on the benchmark calibration results.

The last initial value XT 1 is identified only if we know ρ and λ. It turns out that
to calibrate all three of them in a single system of three equations is feasible. The first
equation of this system is (4.6) for t = 1 which is rewritten here as

1
Œ  1−η
1
‚
NT 1 X T1
= (6.1)
 
N1 ℓ 1 N1 λ
 
(1 − λ)1−λ λλ X 1

where the labor share of the traditional sector in 1650 solves XT 1 given λ since we already
€ Š
have η, N1 , X 1 and ℓ1 from the fitted data. The second equation of the system, again

evaluated at t = 1, solves the output share of the traditional sector. Using the results
derived earlier, we can rewrite this equation as

1
1−η
XT 1
 ‹ η
λ 1−η
YT 1 (1−λ)1−λ λλ X 1
= 1
(6.2)
YT 1 + Y1 1−η
XT 1

N ρb
 € Šλ λ
λ
 ‹ η + (1 − λ) 1 − NT 1 − ℓ 1 N1 1−λ X 1 ℓ0
λ 1−η 1 1
(1−λ)1−λ λλ X 1

69
where b1 , again, is a data point. Finally, the third equation solves the ratio of en-
trepreneurs in adult population from (4.9) for the year 1975 which corresponds to the
14th period of the model:

‚ Œ ‚ Œ
E14 NT 14 ρb14
= (1 − λ) 1 − − (6.3)
N14 N14 ℓ14

The set of (first-stage) data targets of the calibration exercise are thus

NT 1 NT 14 YT 1 E14
2b1 , 2b14 , , , ,
N1 N14 YT 1 + Y1 N14

Using data values of these targets in advance, the unique solution to the system (6.1)-(6.3)
is (numerically) solved such that XT 1 = 0.3951, λ = 0.7102, and ρ = 0.0419.
η
Next, recalling that the initial gross growth rate ζ0 of the traditional sector produc-
1−η
tivity should approximately be equal to n1 for output per worker to stagnate before
the industrial revolution, ζ0 is calibrated to 1.0154 given n1 = s1 b1 from the data and
η = 0.537.

The remaining parameters are θ and π ≡ (σ, ζ1 , ω, φ, γ , ψ). The former is identified

for any value of σ through the normalization of the asymptotic annual growth rate of
productivity to 2.5% given λ. Such a restriction is necessary because the model must
not be featuring a very large asymptotic growth rate that would be counter-factual to
the observed acceleration of the growth rate in the 20th century.29 Returning to the

parameters defining the vector π, a numerical algorithm is used to determine the values
of these five parameters; see below. Clearly, the resulting value of σ returns the value of
θ.

Let md at a ∈ R+d and m(π) ∈ R+d respectively denote the vectors of actual data and

29. Recall that the model predicts that productivity growth accelerates towards the asymptotic equilibrium
given increasing K t and ℓ t .

70
model predictions where d ∈ N++ . The objective function to be minimized is defined as

2
m rd at a − m r (π)

d
X
Q(π) ≡  
r =1 0.5(m rd at a + m r (π))

where m rd at a and m r (π) correspond to r t h dimension of associated vectors, and md at a ∈

R+d and m(π) ∈ R+d vectors are of dimension d = 17 such that the targets include

• log(Output per capita) and log(Population) in 1650, 1750, 1850 and 2000,

• the labor and the output shares of the traditional sector in 1650, 1750, 1850 and
1975, and

• the share of entrepreneurs in 1975.

1650 data targets are included because it is the initial period (t = 1), and 1750 (t = 5)
is, again, the period at which industrial revolution starts. 1850 (t = 9) and 2000 (t =
15) are respectively the periods at which gross fertility is at its historical maximum and

minimum; the data for these periods are informative for parameters (φ, γ , ψ). Finally,
1975 (t = 14) is included because the share of entrepreneurs and the labor and the output
shares of the traditional sector are available for these periods.

The numerical algorithm that returns π minimizes Q(π) under two hidden con-
straints. These constraints that ensure the SGE of period t = 1 is in Regime 1 are

• K1 < K1t r for the invention regime, and

• W1 < (φ + γ )/ℓ1 for the fertility regime.30

The sampling algorithm imfil developed by C. T. Kelley is used to handle these

hidden constraints.31 This algorithm searches the parameter space for a global optima

30. Since a1 = 0, the spillover from the modern to the traditional sector is not active at t = 1. Thus, there
does not exist a hidden constraint for this.

31. The code and the documentation are available online at www4.ncsu.edu/~ctk/imfil.html website.

71
 Table 6.2: The Benchmark Calibration Results

Initial Values Symbol Value

Adult Population N1 2.0718

The Stock of Discoveries K1 1.0000
Mod. Sec. Productivity X1 1.0000
Trad. Sec. Productivity XT 1 0.3951

Parameters Symbol Value

Labor Exponent of the Trad. Tech. η 0.5370

Constant Growth of Trad. Sec. Prod. ζ0 1.0154
Sectoral Spillover Parameter ζ1 1.1877
Labor Exponent of the Mod. Tech. λ 0.7102
Collective Discovery Parameter ω 1.3150
Gross Stepsize of Inventions σ 1.4866
Extrinsic Prod. of Inventive Effort θ 2.3952
"Subsistence" Consumption γ 0.1367
Fertility Preference φ 0.1223
Good Cost of Reproduction ψ 0.0376
Time Cost of Reproduction ρ 0.0419

while ignoring the parameter vectors that violate the hidden constraints. The algorithm
has been modified by randomized initial iterates since the exercises have shown that,
for some initial iterates, the algorithm returns a local optima of Q(π). Specifically, 3
million initial iterates for π have been randomly generated, and around 1 million of these

initial iterates have satisfied the hidden constraints of K1 < K1t r and W1 < (φ + γ )/ℓ1
and the imposed target t t r = 5. The benchmark π is the one that implies the unique
global minimum of the objective function among those obtained for these 1 million or
so random iterates. The benchmark calibration results for all parameters and initial values

are collected in Table 6.2.

72
 log(Output per capita) log(Population) log(Total Output) Labor Share of Trad. Sec.
7 4 12 70
Model
Data 60
3.5
6 10
50

log(millions)
3

percent
40
index

index
5 8
2.5 30

20
4 6
2
10

3 1.5 4 0
1700 1800 1900 2000 1700 1800 1900 2000 1700 1800 1900 2000 1700 1800 1900 2000

Total Fertility Rate Prod. Growth in Modern Sec. Output Share of Trad. Sec. Share of Entrepreneurs
4.5 70 30
gross rate per generation

1.6 60 25
73

4
50
20
3.5

percent

percent
1.4 40
15
3 30
1.2 10
20
2.5 5
1 10

2 0 0
1700 1800 1900 2000 1700 1800 1900 2000 1700 1800 1900 2000 1700 1800 1900 2000

Figure 6.6: Model vs. Data

Notes: Filled circles and squares respectively indicate the model predictions and the actual data described earlier. In the figure that shows
the productivity growth rate of the modern sector, the vertical and the horizontal lines respectively denote the period 1750 and the no-
growth baseline. The model’s predictions for level variables of output are reindexed without any effect on the benchmark calibration
results.
 3 1.6 6
K g (left axis)
t Kt
Ktrt Kt+1 − Kt (right axis)

2.5

(gross growth rate)

1.4 4

1.2 2
1.5

1 1 0
1650 1700 1750 1800 1850 1650 1700 1750 1800 1850 1900 1950 2000

Figure 6.7: The Invention Threshold and the Collective Discovery

Notes: The vertical line on the left panel indicates the period 1750.

6.2 Results

6.2.1 Some Simulations for the Calibration Period 1650-2000

Of prime interest is how the model performs in predicting the observed patterns, and
Figure 6.6 contrasts the model predictions with the actual data for the calibration period.

The model does not perfectly predict the actual data on output per capita, population,

fertility, and so on. On the other hand, even for the variables regarding which the model
performs least successfully, the overall pattern is captured by the benchmark calibration.

What requires a comment is the limited success of the model in predicting output
per capita before mid-19th century. Since the level of population in the model is simply

measured by (1 + b t )N t , the fast increase in total fertility rate 2b t before 1825 leads the
model to predict a declining level output per capita before 1800. On the other hand, total
output is matched more successfully since it is not affected by how the level of population
in the model is defined.

Figure 6.7 pictures the stock K t of useful discoveries and discloses the dynamics of
collective discovery. In the left panel, the decrease of the time-varying threshold K tt r with

74
 4 1.5

Arrival Rate of Inventions

Total Fertility Rate 3.5
1
3
0.5
2.5

2 0
0 500 1000 1500 0 5 10 15
Real Wage Stock of Discoveries

Intrinsic Prod. of Inv. Activity

60 1
Trad. Sec. Labor Share

40 0.8

20 0.6

0 0.4
0 500 1000 1500 0 5 10 15
Real Wage Stock of Discoveries

Figure 6.8: Some Equilibrium Relations

Notes: The model’s predictions for real wage are reindexed without any affect on the
benchmark calibration results.

increasing ℓ t and the expansion of K t are pictured. Naturally, the industrial revolution
starts at 1750 at which K t > K tt r . In the right panel, on the other hand, the gross growth
rate gK t and the mass K t +1 − K t of new discoveries are shown. Increasing E t and ℓ t after
1800 explain the accelerating pace of collective discovery and the decelerating effect of

stabilizing population is observed in the last period.

In Figure 6.8, some equilibrium relations are pictured for the period 1650-2000. The
top left panel show the relationship between real wage and fertility, and the pull effect of
real wage on the traditional sector is observed in the bottom left panel. In the top right

panel, the arrival rate a t of inventions is shown to increase with the stock of discoveries.
The intrinsic productivity ξ (K) of inventive activity, as it is presumed, is shown in the

75
 Entrepreneurship and Invention Entrepreneurs’ Use of Time
2 40
Arrival Rate (left)
Share of Ent. (right)

Invention

(percent)
1 20

Management

0 0
1700 1800 1900 2000 2100 2200 1700 1800 1900 2000 2100 2200

The Decline of the Traditional Sector Society’s Use of Time

80
Labor Share
Output Share
60

Production
(percent)

40

20
Invention
Reproduction
0
1700 1800 1900 2000 2100 2200 1700 1800 1900 2000 2100 2200

Figure 6.9: The Long-Run Evolution of the Equilibrium

Notes: In the right panels, the areas of the figures represent, for each t , the total
available time endowments ℓ t and ℓ t N t .

bottom right panel.

6.2.2 The Equilibrium Path over the Long-Run

The long-run evolution of the model’s static general equilibrium is pictured in Figure
6.9. The top left panel shows how the arrival rate a t of inventions and the share e t of
entrepreneurs in adult population increase during the transition and converge to their
asymptotic values. Importantly, the arrival rate exhibits a logistic shape because of slowly
increasing (K t , ℓ t ) in the early stages of the industrial revolution. The top right and

the bottom right panels respectively show the entrepreneurs’ and the society’s optimal
use of time where the areas of the figures represent, for each t , the total available time

76
 Long−Run Gross Growth Rates (Smoothed)
2

1.5

1
Output p.c.
Real Wage
0.5
1700 1800 1900 2000 2100 2200

1.4
Total Pop.
1.3
Adult Pop.
1.2
1.1
1
0.9

1700 1800 1900 2000 2100 2200

2.5
Mod. Sec. Prod.
2 Trad. Sec. Prod.

1.5

0.5
1700 1800 1900 2000 2100 2200

Figure 6.10: Long-Run Gross Growth Rates (Smoothed)

Notes: The long-run growth rate of each variable is smoothed via robust loess proce-
dure with a span of 7 observations.

endowments ℓ t and ℓ t N t . The transition to the asymptotic equilibrium is characterized

by the decline of management’s share to its asymptotic value while invention’s share, as
its rival, increases. For the society, a large fraction of total available time endowment is
spent on production (including management hours) with a slight decrease in time while

the remarkable trade-off is predicted to occur between reproduction and invention. That
is, the society from pre-industrial times to the modern era finds it optimal to invest more

77
 1.5 100
gKt (left axis)
Kt+1 − Kt (right axis)

1.4 Kt (right axis) 80

(gross growth rate)

1.3 60

1.2 40

1.1 20

1 0
1700 1800 1900 2000 2100 2200

Figure 6.11: The Process of Collective Discovery over the Long-Run

in an intangible "asset", i.e. knowledge, and less in a tangible "asset", i.e. children.

Returning to the long-run growth rates, the model’s predictions for long-run growth

of living standards, population and productivity are shown in Figure 6.10. The smoothed
growth rates indicate the model’s overall success in explaining unified growth phenom-
ena. Further, the lateness of the peak of adult population growth indicates population
aging and the wedge between the sectoral productivity growth rates hints the decline of

the traditional sector once again.

The last figure to be discussed in this subsection, Figure 6.11, pictures the long-run
pattern of collective discovery. As argued earlier, the growth rate gK t converges to its
asymptotic level of unity. While this is happening, the stock of K t itself attains a linear

time-series, converging to ∞, and the total lifetime E t ℓ t of entrepreneurs converges to

the constant E ⋆ ℓ⋆ . Thus, collective discovery continues at an always decreasing rate. On
the other hand, recall that the inventive activity is not affected by how big the stock of
discoveries is for t → ∞.

78
 Survival Probability
1
Benchmark
0.95 Exp.1: µs0=0.6
0.9 Exp.2: µs2=15
Exp.3: µs0=0.6, µs2=15
0.85

0.8

0.75

0.7

0.65

0.6

0.55

0.5
1650 1700 1750 1800 1850 1900 1950 2000

Adult Longevity (Normalized)

0.8
Benchmark
0.75 Exp.1: µl0=0.4165
0.7 Exp.2: µl2=18
Exp.3: µl0=0.4165, µl2=18
0.65

0.6

0.55

0.5

0.45

0.4

0.35

0.3
1650 1700 1750 1800 1850 1900 1950 2000

Figure 6.12: The Experimental Specifications of Exogenous Variables

Notes: See the text for a description of experimental specifications and the results.

6.2.3 Counter-factual Experiments and the Timing Question

Inspired by Desmet and Parente (2009), this subsection presents the results of some
counter-factual experiments on the timing of the industrial revolution. These experi-

ments are important because, once again, the analytical results presented in the previous
chapter do not reveal the overall effects of model inputs on the timing of the industrial

79
 Table 6.3: The Timing Effects of Model Parameters
Parameter Symbol Experiment Industrial Revolution

Fertility Preference φ — No Effect

"Subsistence" Consumption γ 5% Higher Delayed for 1 generation
Time Cost of Reproduction ρ 25% Higher Delayed for 2 generations
Good Cost of Reproduction ψ 10% Higher Delayed for 1 generation

Collective Discovery ω 50% Lower Delayed for 2 generations

Labor Exp. of Mod. Tech. λ 10% Lower Delayed for 6 generations
Stepsize of Inventions σ 25% Lower Delayed for 3 generations
Extrinsic Prod. of Inv. Effort θ 50% Lower Delayed for 6 generations

Con. Gr. of Trad. Sec. Prod. ζ0 5% Higher Delayed for 1 generation

Sectoral Spillover Parameter ζ1 — No Effect
Labor Exp. of Trad. Tech. η 25% Lower Delayed for 1 generation

revolution due to the second-order effects. Besides, building on the benchmark calibra-
tion results, the timing question can be approached specifically for the England’s indus-
trial revolution. The results, however, should still be read only as suggestive because of

the simplicity of the model economy and the "ceteris paribus" assumption behind the
experiments.

The first set of experiments investigates the role of differing survival probability (st )
and adult longevity (ℓ t ) levels. For each exogenous variable, some of the estimated pa-
rameters of logistic fits are altered ceteris paribus. Specifically, for st , (i) µ s 0 is decreased
from 0.63 to 0.6 in Experiment 1, (ii) µ s 2 is increased from 12.83 to 15 in Experiment

2, and (iii) Experiments 1 and 2 are merged to design Experiment 3. Similarly, for ℓ t ,
µℓ0 is decreased by 0.01 units to 0.4165 and µℓ0 is increased to 18. Figure 6.12 shows the
resulting experimental inputs.

The results indicate that even such small changes in st and ℓ t create effects on the
timing of the industrial revolution. For st , Experiments 1 and 3 imply that the industrial
revolution is delayed by one generation, starting at 1775, but no hastening or delaying

effect is found in Experiment 2. For ℓ t , Experiment 1 and 2 also imply a one-generation

delay, but the delay is equal to three generations when these changes merged in Experi-

80
ment 3.
The second, the third and the fourth sets of experiments respectively investigate the
effects of (i) (φ, γ , ρ, ψ) that directly affect population growth, (ii) (ω, λ, σ, θ) of inven-
tion and collective discovery technologies and (iii) (ζ0 , ζ1 , η) that governs the productivity

growth and the size of the traditional sector. Table 6.3 summarizes the results.
First recalling that fertility preference φ does not have threshold and growth effects
when the industrial revolution occurs before the fertility decline, the timing of the indus-
trial revolution is not affected by φ under the benchmark calibration. Next note that the
sectoral spillover parameter ζ1 does not create timing effects as expected. Another note-

worthy result is that the hastening growth effects of parameters γ and ψ are dominated
by the second-order effects on population growth. That is, when γ and ψ are higher than
their benchmark levels, ceteris paribus, the associated decreases in the rate of population
growth before the industrial revolution decreases the growth rate gK t of useful discoveries

with N t /K t now being allowed to change. Finally, it should be noted that the ambigu-
ity of the effect of λ is resolved such that the threshold effect dominates the supply of
entrepreneurship effect. That is, when λ is lower than its benchmark level, there exist
more entrepreneurs that collectively discover but invention becomes optimal later than

the benchmark due to the smaller contribution of inventions to profits.

Overall, the counter-factual experiments show that small deviations from the bench-
mark model create large timing effects. Even if the model is not applied as a prototype
model of the industrial revolution to economies other than England, the counter-factual
experiments suggest that gains in adult longevity or the quality of collective discovery,

e.g., might have played non-trivial roles in the timing of England’s transition to modern
economic growth. The bigger questions such as whether the first industrial revolution
did not occur in China really because of the low quality of collective discovery there or
whether the first industrial revolution could not have occurred in Sub-Saharan Africa

really because of its disease environment remain open.

81
 Chapter 7

Discussion

This chapter provides brief discussions on some aspects of the model and shows that
the model can be extended in ways that may provide further insights on the role of en-
trepreneurship and knowledge for the Industrial Revolution.

7.1 The Industrial Revolution: Break or Continuity?

The unified growth model studied above has a stark implication: Entrepreneurs of a
"special" generation find it optimal to direct resources into risky inventive activities un-
like those of past generations. These entrepreneurs are "special" because the number of

useful discoveries they have access to, given their longevity, is large enough to signal a
higher expected level of profit for them if they are to decrease the time they could spend
to routine management. In a sense, they benefit from standing on the shoulders of dead
entrepreneurs who collectively created all these useful discoveries in a serendipitous way.

The invention threshold in the model leads to a kinked time-series of labor productiv-
ity in the modern production sector, and this in turn implies a kinked time-series of the
real wage that exhibits exponential growth starting with the industrial revolution. The
Industrial Revolution in history is matched by an invention revolution in the model. Af-
ter this invention revolution, exerting inventive effort to appropriate an increasing profit

remains optimal throughout the history. Figure 7.1 shows the data collected by Sullivan
(1989) on the number of process innovations patented in England.
 Patents Granted for Process Innovations in England

2
10
(Logarithmic Scale)

1
10

Pre−1750 Avg. = 5.87

0
10
1660 1680 1700 1720 1740 1760 1780 1800 1820 1840

Figure 7.1: Patented Process Innovations in England, 1661-1850

Whether the first Industrial Revolution, roughly covering the period from 1760 to
1830, is a break from the past or a continuity has remained controversial among some
economic historians. The gradualist view of Crafts and Harley (1992) suggests that there

was little economic growth in England until the early 19th century in per capita terms
and that the scope of fast technological progress was limited with the textile sector before
the diffusion of the steam technology. The first argument has later been advanced by
new estimates of Clark (2001), yet the notion of an industrial revolution as a structural

break characterized by very slow growth in per capita terms is not controversial at all.32
Studied extensively by Pereira (2003), several variables of interest, including total indus-
trial output and population, exhibit endogenously determined upward breaks during the
first Industrial Revolution, and Mokyr (2004) and others suggest that what kept output

32. As noted by Crafts’s (2005, p. 533), "[g]radualism in the transition to modern economic growth should
not be confused with an absence of fundamental change."

83
per capita at a very low level during the first Industrial Revolution was indeed the fast
33
expansion of English population.

The model economy constructed and studied above, as a unified model, captures ex-

actly this type of dynamics between population and technology. The predicted time-
series of output per capita does not exhibit growth for a couple of generations after the
start of the industrial revolution, and increasing output due to endogenous technolog-
ical progress hardly overcomes the population pressure until the pace of technological

progress becomes fast enough.

The model economy’s industrial revolution thus fits well with the notion of a struc-
tural break without conflicting with the view that a sort of continuity with the past exists.
In essence, the period at which the industrial revolution in the model starts is determined

by the latent dynamics of the model economy before the industrial revolution.

7.2 Serendipitous Inventions

That the rate of technological progress in the modern production sector before the Indus-
trial Revolution is zero is counter-factual to what we observe in the data: As noted earlier,

the real wage series in England has an upward trend after mid-1600s, and a minuscule rate
of growth in the real wage before the Industrial Revolution is also consistent with the
patent data of Sullivan (1989): The number of patented process innovations per year, al-
beit being trendless, implies minor improvements in productivity and hence in the real

wage. A question of interest is thus whether the model can be extended to account for
such haphazard type of technological progress.

The simplest extension along this line of thought is to allow for serendipitous in-

ventions to exogenously increase the baseline productivity X t of the modern sector.

33. Temin (1997), attacking the second argument of the gradualist view, shows that England was a net ex-
porter not only in cotton textiles and iron goods sectors where technological progress was fast but also
in many other industries. Crafts and Harley (2000), in defense, use a computable general equilibrium
model that shows exports may increase in the absence of sectoral technological progress. Pereira (2003)
rightly argues that those of Temin (1997) and Crafts and Harley (2000) were indirect tests.

84
Serendipitous inventions can be thought of as resulting exogenously without altering the
optimal behavior of entrepreneurs regarding the inventive activity. The law of motion
for X t can simply be extended to include serendipitous inventions as in

X t +1 = X t exp[(σ − 1)(a t + a s )]

where a s > 0 represents the arrival rate of serendipitous inventions under the additional

assumption that a serendipitous invention has the same stepsize σ > 1 of a purposeful
invention. Clearly, whenever a t = 0, the gross growth rate of X t reduces into exp[(σ −
1)a s ].

If allowed to be a positive number, the arrival rate a s of technological progress through

serendipitous inventions would affect the timing and the pace of demographic transition
and structural transformation through the real wage channel. The qualitative nature of
the model’s analytical results, most notably the existence of an invention threshold, nev-

ertheless remain same under this simple extension.

7.3 Managers vs. Inventors

One legitimate concern might be over the presumption that entrepreneur and inventor
is the very same individual in the model. This presumption, recalling the motivating
evidence by Meisenzahl and Mokyr’s (forthcoming), is the simplest way to let inventors
be incentivized via profit motive within the occupational choice framework adapted.

In a simple alternative formulation in which inventors are still incentivized by profit

shares, entrepreneurs spend their entire labor endowment to management and make con-

tracts with "freelance" inventors for the latter to undertake the inventive activities. Under
certain simplifying assumptions, the invention threshold property of the basic model is
preserved.

A "freelance" inventor is in essence a worker who may find it optimal to spend some
of her labor endowment to inventive activities. To completely assume away search and

85
matching frictions previously emphasized by Michelacci (2003), suppose that an inventor
is always matched with an entrepreneur. To simplify the matters even more, let the con-
tract between entrepreneur i and inventor i be such that an exogenously given fraction
1 − ν t ∈ (0, 1) of firm i’s ex post profit is appropriated by inventor i.34 This variable can

thus be argued to represent a dimension of the quality of innovation-promoting institu-

tions in period t as in Jones (2001).

Inventor i’s expected (lifetime) earning under these assumptions can be written as

 Γ !
Xt ai t
(1 − ν t ) exp Σai t Λ  ℓt −


 ℓt + Wt

Wt θξ K t

where the first term indicates the expected profit obtained by inventor i and the second
term is her wage income. Since fertility choice by inventor i is still separable from the
choice of ai t , optimal inventive effort is zero if

h  λ i−1
(1 − ν t )ξ K t ℓ t < θ σ − 1


1−λ

and the start of an industrial revolution not only requires a large enough stock of discov-
eries and a high enough adult longevity but also depends crucially on whether the society
sufficiently rewards its potential inventors.35

7.4 The Fishing Out Effect

One noteworthy aspect of productivity growth in the simple version of the model is
the absence of the so-called fishing out effect: The (unit) productivity of labor directed
to inventive activity does not change and, hence, not decrease with the level of baseline

34. Ideally, ν t could be allowed to follow endogenously from a bargaining problem between the en-
trepreneur and the inventor as in Michelacci (2003). A closed-form solution to the model in this case,
however, does not exist.

35. Note that the unique SGE of this version of the model  still features occupational choice through

 X t ‹Γ ai t
ν t EΠi t = W t ℓ t = (1 − ν t ) exp Σai t Λ W ℓt + Wt ℓt −
t θξ (K t )

86
productivity X t . No matter how high X t is, the arrival rate ai t of inventions is fixed for
given levels of K t and H r i t . The endless expansion of K t makes inventive effort always
more productive in time as postulated, but the only limit realistically imposed is on the

level of this epistemic effect when K t → ∞.

To incorporate the fishing out effect by making the arrival rate ai t a decreasing func-
tion of X t , the productivity term ξ (K t ) of the model can be redefined as ξ (K t , X t ) which
satisfies
∂ ξ (K t , X t )
< 0 for all K t
∂ Xt

Here, it is again required that, for any X t < ∞,

∂ ξ (K t , X t ) € Š € Š
> 0 for all K t < ∞ ξ 0, X t = 0 lim ξ K t , X t = 1
∂ Kt Kt →∞

Not surprisingly, the invention threshold and the asymptotic equilibrium of the
model would be affected from the presence of the fishing out effect: A higher initial
level X 0 of modern sector productivity unambiguously delays the start of the industrial

revolution. This, in a sense, is a technological lock-in result and is itself interesting as a

theoretical possibility, but the hypothesis of induced innovation, applied to the Industrial
Revolution by Allen (2011), weakly rejects this possibility to be true: The real wage in
urban areas of Britain was higher in comparison to those of other European cities.

Returning to the asymptotic equilibrium, the fishing out effect may lead the asymp-
totic growth rate of modern sector productivity and output per capita to converge to

zero. To be specific, let ξ (K t , X t ) be defined as

1
ξ (K t , X t ) ≡ 1 − Kt
1+
Xt

which is possibly the simplest functional form that satisfies the restrictions stated above.
The advanced stages of economic development under this scenario are characterized by

87
the decline of the knowledge-productivity ratio

Kt
kt ≡
Xt

to its invention threshold

1
k tt r ≡  λ 
θ σ 1−λ − 1 ℓ t − 1

because the (gross) growth rate gK t of the stock of discoveries again converges to unity
with the maximum level N ⋆ of adult population being constant.36 The model therefore
asymptotically behaves like the semi-endogenous (unified) growth model of Jones (2001)
and the model of Strulik and Weisdorf (2008) due to the fishing out effect. The actual

very long-run growth rate of income per capita on the other hand does not exhibit a
downward trend for the developed economies. Overall, the implications of the fishing
out effect seem to be contradicting with observed patterns.

7.5 The Scale Effect in the Process of Collective Discovery

Since the mass E t of entrepreneurs is proportional to adult population N t , the process

of collective discovery is characterized by a scale effect. Specifically, given the share e t of
entrepreneurs in adult population, the growth rate gK t of the stock of useful discoveries
is increasing in N t :

 
N ρb
ωℓ t E t ωℓ t (1 − λ) 1 − NT t − ℓ t N t
t t
gK t = 1 + =1+
Kt Kt

This scale effect may raise the question of why economies that had bigger populations
in pre-industrial era compared to England, e.g., China, did not achieve the first industrial

revolution even though, as argued in Chapter 5, such questions should be formulated

with great care. If taken seriously, still, an answer to the question is provided by the

36. Note that the asymptotic equilibrium thus characterized is globally stable.

88
determinants of collective discovery other than N t : It is not solely the mere totality of
adult hours ℓ t N t that explains the growth rate of K t but also (i) the quality ω of the
process of collective discovery and (ii) the share e t of entrepreneurs in adult population.

As noted earlier, ω as a structural parameter represents the quality of the environ-

ment in which entrepreneurs create and share useful discoveries. England here had the
advantage of being a small country with respect to (absolute) geographical size. Also

advantages of England, as argued by those stressing the role of collective discovery and
industrial enlightenment, are (i) the gentlemanly behavior and the technological motiva-
tion of business owners and (ii) the efficiency of social networks and informal institu-
tions. A sufficiently large ω for England may well have dominated the negative effect of
its comparatively small population.

Returning to the share of entrepreneurs, first note that pre-industrial fertility lev-
els around the world were not significantly different. If one further assumes that pre-

industrial longevity levels and parameters λ and ρ were similar in England and elsewhere,
the prime determinant of the share of entrepreneurs would be the size of the traditional
sector. The limited data here indicates that England in pre-industrial times had a higher
rate of urbanization than China; see Voigtländer and Voth (2006).

In general, any rival use of time endowment is important in determining the supply
or, put more correctly, the lack of entrepreneurship, and the labor shares of occupations

that do not contribute to collective discovery would have delaying growth effects for the
timing of the industrial revolution.

One such occupation regarding which England had arguably an advantage compared
to China is state bureaucracy. Imagine, first, a richer framework with a political authority
where bureaucrats are employed by the state to produce a public good. A larger state
bureaucracy in this case would decrease the growth rate of K t , ceteris paribus, because the
mass of entrepreneurs who collectively discover would be smaller. This is true (i) whether

the bureaucracy is financed through distortionary taxation or not and (ii) whether the
productivity of a unit hour of a bureaucrat in producing the public good in question is

89
increasing or stagnant. Obviously, if the bureaucrats work with stagnant productivity,
sustained growth in social welfare is not possible before an industrial revolution, and a
small state bureaucracy is a desired feature to increase the supply of entrepreneurship
ceteris paribus. England might indeed have benefited from avoiding a large professional

bureaucracy, as noted by Mokyr (1998), and China’s potential might indeed have been
restricted by its large and ineffective bureaucracy, as emphasized by Landes (2006).

7.6 Science and Scientists

The anecdotal evidence show that professional scientists’ direct contribution to the first

Industrial Revolution was limited. Also well-known is that British inventors, compared
to those of other European nations, were particularly successful in applied sciences that
built heavily upon the abstract contributions of, e.g., German and French scientists. That
there does not exist a strong causality running from scientific progress to an industrial
revolution is also supported by the fact that neither China nor the Islamic civilization,

both scientifically superior to Europe at certain eras of antiquity, did realize an earlier
industrial revolution. All these, together with the lack of reliable data on the number of
scientists and a useful theoretical framework of the economics of science, motivate the
model to exclude the role of science and scientists for the process of collective discovery.

A unified growth model that exploits the nexus between discoveries and inventions
might nevertheless be expected to incorporate the role of science and scientists. One of

the most important actor of the story of technological progress after the first Industrial
Revolution is surely the professional scientist, and three questions, at least, remain open
for the unified growth theory. First, why and how the grant-like forms of science pa-
tronage dominated the prize-like forms of it starting with the 18th century, a pattern
documented by Hanson (1998), is central to the rise of professional scientists. Second,

as emphasized by Pumfrey and Dawbarn (2004), science patronage exhibited a historical

transition from being mostly ostentatious to being mostly utilitarian, starting first in the

90
16th century England. Finally, there does not exist a formal economic theory of the Sci-
entific Revolution which was another major discontinuity in the history of mankind. A
network model of the Scientific Revolution, once hinted by Kelly (2005), may be a use-
ful starting point to understand the formation and the movement of European scientist

networks between 1400 and 1900 documented by Taylor et al. (2008).

7.7 Endogenous Survival and Longevity

The model restricts the knowledge content of useful discoveries with the knowledge of
natural phenomena that are related solely with the production processes. In a richer
framework with scientists, discoveries could be thought of being applicable also to
medicine. This second role of useful knowledge is exactly what Easterlin (1995) sug-

gests; the Mortality Revolution of Europe did crucially depend on the creation and the
diffusion of useful medical knowledge just as the Industrial Revolution benefited from
the advances in physics, chemistry, geology, and other fields. In England, for example,
the number of published books on health grew 9-fold from 1600 to 1800 as noted by

de la Croix and Sommacal (2009).

In addition to the models that postulate a knowledge-mortality link via human capi-
tal accumulation, e.g. Cervellati and Sunde (2005) and Lagerlöf (2003), the conjecture has
been formalized by Mokyr (1993) and more recently within unified growth frameworks
by de la Croix and Sommacal (2009) and Strulik and Weisdorf (2011). Note that endo-

genizing survival probability (st ) and adult longevity (ℓ t ) in a richer framework with
scientists would pose no serious analytical difficulties. Given that (i) the stock of useful
discoveries would still be growing along the transition and (ii) st and ℓ t would still have
the logistic shapes fitted against time in the present formulation, the only complication

is to calibrate the parameters of two logistic functions of K t , one for st and the other for
ℓt .

91
 Chapter 8

Concluding Remarks

A transition characterizes the economic history of today’s developed economies. This is

a transition from stagnating to growing levels of material living standards. It involves the
declines of mortality and fertility rates with a bell-shaped pattern of population growth

through time, i.e. the demographic transition. (Adult) longevity, the participation to and
the length of formal education, the openness to trade, and the levels of urbanization and
industrialization all increase along this transition. Democracy, the last but not the least,
becomes persistent in the developed world where the transition had first started. In short,

this is the transition to what social scientists call modernity. This, in Clark’s (2007, p. ix)
words, is big history.

The turning point of big history was the Industrial Revolution which represents

a structural break in the sense that technological progress was no longer simply due
to serendipitous inventions. Schumpeter’s (1934) "entrepreneur-inventor"s, seeking in-
creased market shares and profits, took the stage instead, and the world was not the same
when the first corporate R & D lab was opened by Thomas Edison in 1876.

This dissertation studies a view of the Industrial Revolution that promotes the dual
role of entrepreneurship for inventions and discoveries; the serendipitous expansion of
the latter eventually leads to purposeful activation of the former. No such thing as an

industrial revolution occurred for a very long episode of history because not enough was
known about natural phenomena and lives were very short. Yet the type of useful knowl-
edge relevant to production processes was created by and diffused among entrepreneurs.
In one sense, it had to be because they were managing the firms utilizing these production
processes.
The simple unified growth model of this mechanism constructed in this dissertation

leaves many questions, other than the ones discussed in the previous chapter, open for
further research. First, the emphasis is biased on the supply-side determinants of inven-
tive activity, and the roles of market size and demand remain implicit within the two-
occupation general equilibrium framework. How inventions are brought to markets as
innovations is no less an important question. Second, the model simply features perfectly

competitive innovation, and the role of patents for the industrial revolution, still contro-
versial among economic historians, is ideally to be incorporated within a richer treat-
ment. Third, the effects of knowledge diffusion on European and global scale, through
the mobility of goods and people, is yet to be explored in a unified growth framework.

The simple model of this dissertation may serve as a starting point of such an explo-
ration. The last but not the least, how the rises of democracy and formal education are
interrelated with the enlightenment of the economy through useful knowledge is an open
question.

93
 Appendix A

Proofs

Proof of Lemma 1:

(3.17′ ) can simply be rewritten as

λ λ
∞ a z σ ( 1−λ ) z
  1−λ !
λ Xt ai t X it
EΠi t = exp −ai t (1 − λ)λ ℓt −


1−λ  

Wt θξ K t z=0 z!

after some arrangements. By Taylor’s Theorem, the summation term on the right is
 λ 
identical to exp σ ai t . Thus, we have
1−λ

 λ
 1−λ !
 λ
 λ Xt ai t
EΠi t = exp σ 1−λ ai t − ai t (1 − λ)λ 1−λ   ℓt −

Wt θξ K t

λ λ
Defining and substituting Γ ≡ λ
1−λ
, Λ ≡ (1 − λ)λ 1−λ , and Σ ≡ σ 1−λ − 1 yield (3.17′′ ) and

complete the proof. „

Proof of Proposition 1:

The uniquely existing SGE follows from three features of the model:

1. Both decision problems have unique solutions since the objective functions are
strictly quasi-concave and differentiably continuous on compact choice sets.

2. At these unique solutions, a unique level of real wage makes individuals indifferent
between becoming an entrepreneur and becoming a worker.

94
 3. This unique level of real wage is also the one that clears the labor market and, hence,
that (residually) determines the mass of entrepreneurs.

Starting with the workers’ problem characterized by (3.23) and (3.24), the Kuhn-

Tucker F.O.C.s imply


 <0
 if nw t = 1
1
‚ Œ
∂ Uw t φ  
(W t ℓt −γ )s t
=−


ρW t + ψ + >0 if nw t =
∂ nw t st nw t 


ρW t +ψ

=0 otherwise


Thus, optimal net fertility by workers satisfies


(W t ℓt −γ )s t φ+γ

 ρW t +ψ
if W t < ℓt

 h i
φs t φ+γ φs t −ψ
nw t = if W t ∈ , ρ (A.1)

 ρW t +ψ ℓt

φs t −ψ
1 if W t >

ρ

Next note that the entrepreneurs’ problem characterized by (3.26)-(3.28) is separable

in ni t and ai t . With respect to the latter, we have

 Γ 
∂ EUi t Xt 1
= exp Σai t Λ   −



+
∂ ai t Wt θξ K t

!
ai t  <0 if ai t = 0
Σ ℓt −
 (A.2)
θξ K t
€ Š
=0 if ai t ∈ 0, a tmax


Hence, there always exists a unique solution ai t ≥ 0.37 Moreover, this solution is symmet-
ric, i.e. ai t = a t ≥ 0 for all i, since all entrepreneurs face the same set of given variables.
In turn, the expected profit EΠi t is unique as well:

€ ŠΓ
EΠi t = exp Σa t Λ X t /W t ℓ t − a t /θξ K t (A.3)

37. Notice that (i) the S.O.C. for maximum is satisfied at this solution, and (ii) ai t = a tmax is never optimal
because it implies zero profits.

95
 With respect to fertility ni t , the Kuhn-Tucker F.O.C.s imply


 <0
 if ni t = 1
1
‚ Œ
∂ EUi t
φ  
(EΠi t −γ )s t
=− ρW t + ψ + >0 if ni t =
∂ ni t st ni t 


ρW t +ψ

=0 otherwise


Therefore, optimal net fertility by entrepreneurs satisfies


(EΠi t −γ )s t

 ρW t +ψ
if EΠi t < φ + γ


φs t φs t −ψ
ni t = if EΠi t ≥ φ + γ and if W t ≤ (A.4)

 ρW t +ψ ρ

φs t −ψ
1 if W t >

ρ

Lemma A.1: In the unique SGE, we have nw t = ni t = nt .

Proof — The intuition behind this lemma is twofold: First of all, the total cost of re-
production in terms of gross fertility, denoted by ρW t + ψ, is identical to workers and
entrepreneurs. Second, entrepreneurs and workers, in equilibrium, are forced to derive
equal (expected) utilities, i.e. EUi t = Uw t .

Formally, the equal utilities restriction can be rewritten as

‚ Œ ‚ Œ

ni t
nw t
EΠi t − ψ + ρW t + φ ln ni t = W t ℓ t − ψ + ρW t + φ ln nw t (A.5)



st st

In what follows, nw t is taken as a benchmark, and it is shown, for each characteriza-

tion of nw t , that nw t 6= ni t leads to a contradiction.

(W t ℓt −γ )s t
• The Case of nw t = ρW t +ψ
:

In this case, we must have from (A.1) that

φ+γ
Wt < (A.6)
ℓt

φs t −ψ
– ni t = 1 in this case leads to a contradiction because (A.4) implies W t > ρ
.

96
 φs t φs t −ψ
– ni t = ρW t +ψ
requires EΠi t ≥ φ + γ and W t ≤ ρ
from (A.4), and (A.5) now
reads

(W t ℓ t − γ )st
‚ Œ ‚ Œ
φst
EΠi t − φ + φ ln = γ + φ ln
ρW t + ψ ρW t + ψ
Wt ℓt − γ
‚ Œ
EΠi t − (φ + γ ) = φ ln
φ

Note from (A.6) that the R.H.S. must be a negative number. On the other
hand, EΠi t ≥ φ + γ implies that the L.H.S. is non-negative. Hence, we have a
contradiction.

(W t ℓt −γ )s t
– The remaining task for the case of nw t = ρW t +ψ
is to make sure that EΠi t =
(W t ℓt −γ )s t
W t ℓ t when EΠi t < φ + γ . Re-writing (A.5) for nw t = ρW t +ψ
and ni t =
(EΠi t −γ )s t
ρW t +ψ
confirms this:

(EΠi t − γ )st (W t ℓ t − γ )st

‚ Œ ‚ Œ
γ + φ ln = γ + φ ln
ρW t + ψ ρW t + ψ
(EΠi t − γ )st (W t ℓ t − γ )st
‚ Œ ‚ Œ
ln = ln
ρW t + ψ ρW t + ψ
EΠi t = W t ℓ t

φs
• The Case of nw t = ρW +ψ
t
:
t

In this case, we must have from (A.1) that

φ + γ φst − ψ
– ™
Wt ∈ ,
ℓt ρ

φs t −ψ
– ni t = 1 again leads to a contradiction because (A.4) implies W t > ρ
.

97
 (EΠi t −γ )s t
– ni t = ρW t +ψ
requires EΠi t < φ + γ . (A.5) in this case reads

(EΠi t − γ )st
‚ Œ ‚ Œ
φst
γ + φ ln = W t ℓ t − (φ) + φ ln
ρW + ψ ρW t + ψ
‚ t
EΠi t − γ
Œ
φ ln = W t ℓ t − (φ + γ )
φ

Once again, we have a contradiction because the L.H.S. is negative and the

R.H.S. is non-negative.

φs t
– Note that ni t = ρW t +ψ
as in the previous case confirms EΠi t = W t ℓ t via (A.5).

• The Case of nw t = 1:

In this case, we must have from (A.1) that

φst − ψ
Wt > (A.7)
ρ

φs t φs t −ψ
– ni t = ρW t +ψ
leads to a contradiction because (A.4) implies W t ≤ ρ
.

(EΠi t −γ )s t
– ni t = ρW t +ψ
requires EΠi t < φ + γ . (A.5) in this case reads

(EΠi t − γ )st 1
‚ Œ ‚ Œ
γ + φ ln = W t ℓ t − ψ + ρW t + φ ln (1)


ρW t + ψ st
 
−
‚ Œ
EΠ
 it γ ψ + ρW t
φ ln  ρW +ψ  = W t ℓ t − γ −

t st
st
ℓ t st − ρ
‚ Œ ‚ Œ
γ st + ψ
= Wt −
st st
ℓ t st − ρ
‚ Œ‚ Œ
γ st + ψ
= Wt −
st ℓ t st − ρ

ρW t +ψ
Since (A.7) implies st
> φ, the L.H.S. of this last equation is negative

98
 given EΠi t < φ + γ . The R.H.S. is on the other hand positive because

φst − ψ φ+γ
Wt > >
ρ ℓt

where the second inequality implies

ψℓ t + ργ
φ>
ℓ t st − ρ

ψℓt +ργ
Since we have W t ℓ t − γ > φ as well, W t ℓ t − γ > ℓt s t −ρ
implies that W t >
γ s t +ψ
ℓt s t −ρ
.

– Finally, note that ni t = 1 as in the previous cases confirms EΠi t = W t ℓ t via

(A.5).

Hence, we conclude that the equal utilities restriction EUi t = Uw t is satisfied with
EΠi t = W t ℓ t and nw t = ni t = nt . ƒ

To proceed, notice that EΠi t = W t ℓ t and (A.3) solve W t given the unique solution
a t ≥ 0 that follows from (A.2). Next, given a t , (3.9) implies h r i t = h r t , and (3.18) then
returns h mi t = h mt for the given level of ℓ t . Given W t , on the other hand, the unique

SGE levels of HT t , NT t , YT t follow respectively from (3.20), (3.21) and (3.12′ ). Note

that the realizations of stochastic variables zi t and Xi t simply follow from (3.8) and (3.7),
respectively. Given productivity Xi t , (3.15), (3.16) and (3.5) respectively solve the unique
SGE levels of hw i t , Πi t , Yi t . (3.22) solves Cw t , and (3.25) solves Ci t . Thus, only E t and

Yt remain to be solved.

What solves E t is the labor market clearing condition (3.30). To see this, first recall
that the arrival rate a t is common across entrepreneurs. This and the fact that invention
events are independent across entrepreneurs imply, via (Borel’s version of) the law of

large numbers, that the ex post fraction of entrepreneurs with z ≥ 0 inventions for any
a z exp(−a t )
given a t is equal to the ex ante Poisson probability t z! of achieving z ≥ 0 inventions.

99
This property allows us to write

ZEt ∞
X

a tz exp −a t


hw i t di = E t   hw t (z) (A.8)
z=0 z!
0

where hw t (z) reads, as implied by (3.15),

 λ 
1−λ
!
1 λ X a
hw t (z) ≡ λ 1−λ σ ( 1−λ )z 
 t

 ℓ − t
1  t

 1−λ
Wt θξ K t

Applying the reasoning of Lemma 1 to the R.H.S. of (A.8) implies

 λ 
ZEt 1−λ
!

1  Xt  at
hw i t di = E t exp Σa t λ 1−λ   ℓ −
1  t

 1−λ
Wt θξ K t
0

This last equation and (3.30) then solve E t .

Finally, given E t , Yt is solved from (3.19) using the above reasoning

ZEt
a tz exp −a t


∞
X
Yi t di = E t   Yt (z)
z=0 z!
0

 λ 1

λ
where Yt (z) satisfies Yt (z) ≡ λ
λ
1−λ (
σ 1−λ ) z 1−λ
X t /W t 1−λ
ℓ t − a t /θξ K t . „

Proof of Proposition 2:

The invention threshold simply follows from (A.2). The (symmetric) solution of a t is at
boundary, i.e. a t = 0, if

 Γ  
Xt 1
Λ  −
+ Σℓ t  < 0
Wt θξ K t

100
 € ŠΓ
Since Λ > 0 and X t /W t > 0, this inequality implies

h  λ i−1
ξ K t ℓ t < (θΣ)−1 = θ σ 1−λ − 1

 λ −1
The interior solution a t = θξ K t ℓ t − σ 1−λ − 1 > 0 follows, again, from (A.2). „

Proof of Proposition 3:

As the proof of Lemma A.1 shows, the expected profit EΠi t in the unique SGE is equal
to W t ℓ t . Hence, (A.1) and (A.4) imply the desired result given nw t = ni t = nt . Further-
more, since st is common across entrepreneurs and workers, we have bw t = bi t = b t =

nt /st . „

Proof of Proposition 4:

The existence and the uniqueness of period-t SGE and that the laws of motion for en-
dogenous state variables, i.e. (3.1), (3.13), (4.11) and (4.12), are all one-to-one functions
imply the existence and the uniqueness of the DGE for the entire history from t = 0 to
t → ∞. „

Proof of Proposition 5:

First note that ℓ t E t > 0 for all t implies K t +1 − K t > 0 for all t . Next, it is assumed that
K tt r < ∞. Thus the continuing growth of the stock K t of useful discoveries eventually
makes the inventive activity optimal at some period t t r where K t t r ≥ K tttrr . „

101
 Appendix B

The Conditional Dynamical System

As noted earlier, the dynamical system of the model cannot be rewritten as a simple
autonomous system of normalized variables due to its severe non-linearity. A formal
discussion of global stability however is still feasible since the closed-form solution of
the model’s unique SGE allows us to study the qualitative properties of a conditional
dynamical system with a phase diagram. The modest purpose of this appendix is to show

why the asymptotic equilibrium is globally stable.

The conditional dynamical system to be considered, possibly the simplest among all,
is that of (X t , q t ) where we define q t as follows:

Nt
qt ≡
Kt

The asymptotic equilibrium on (X t , q t ) plane is now characterized by X t → ∞ and q t →

0 for t → ∞.

This system is conditional on endogenous state variables (N t , K t , X t , XT t ) and exoge-

nous state variables (st , ℓ t ). For notational ease, define Zt ≡ (N t , K t , XT t , st , ℓ t ) as the

vector of state variables excluding X t . In what follows, the notation is slightly abused by
redefining some of the model variables as functions of Zt even if some elements of Zt do
not affect the variable in the question.

The first equation of the (X t , q t ) system governs X t and rewritten here as

X t +1
= exp (σ − 1) a(Zt )
 
Xt

102
 qt

X Xt

q̂ t

0 Xt
0

Figure B.1: The X X t Set and the Dynamics of X t

where the arrival rate a(Zt ) is as in (4.1). On (X t , q t ) plane, the set of vectors satisfying

X t +1 = X t is clearly

Nt Nt
q t ≥ q̂ t ≡ q̂(Zt ) ≡ = −1 (XX t )
K tt r
•  ˜
λ
−1 −1
ξ θ σ 1−λ −1 ℓ−1
t

and X t grows wherever q t < q̂ t , i.e. wherever, for any given value of adult population N t ,

the stock of discoveries is sufficiently large (K t > K tt r ). Importantly, for any given value
of X t , q̂ t is increasing in N t and ℓ t . Figure B.1 pictures the dynamics of X t regardless of
how q t changes in time.

The second equation of the system, the one that governs q t , reads

q t +1 N t +1 /N t n(Zt , X t ) n(Zt , X t )
≡ = =  ‹
qt K t +1 /K t 1 + ωℓ t e(Zt , X t )q t 1 + ωℓ t f N M (Zt , X t ) −
ρn(Zt ,X t )
qt
s t ℓt

where n(Zt , X t ) = nt is the level of net fertility, e(Zt , X t ) = E t /N t is the share of en-

103
trepreneurs in adult population, and f N M (Zt , X t ) = 1 − (NT t /N t ) is the labor share of
the modern sector. The locus of vectors satisfying q t +1 = q t on (X t , q t ) plane is thus

n(Zt , X t ) − 1
qt =  ‹ (qq t )
ρn(Zt ,X t )
ωℓ t f NM
(Zt , X t ) − s t ℓt

As usual, the shape of q q t locus on (X t , q t ) plane and the dynamics of q t below and
above this locus are of interest. Regarding the latter, we simply have ∂ (q t +1 /q t )/∂ q t < 0

for all q t . Thus, q t increases wherever q t is below the q q t locus and it decreases otherwise.

The shape of the q q t locus is determined by f N M (Zt , X t ) and n(Zt , X t ). Regard-

ing the former, we know that f N M (Zt , X t ) is a strictly increasing function of X t for
any (Zt , X t ) due to the decline of the traditional sector with growing X t . Net fertility

n(Zt , X t ), on the other hand, changes with X t non-monotonically since there exist three
regimes of net fertility determined by X t given Zt . These regimes are separated by two
thresholds of X t such that

 1
∂ n(Zt , X t ) ˆI  φ+γ λ

>0 if X t < X t ≡

∂ Xt ℓ t (1 − λ)1−λ λλ δ a t , K t , ℓ t
 1
ˆ II  φst − ψ λ

n(Zt , X t ) = 1 if X t > X t ≡ 
ρ(1 − λ)1−λ λλ δ a t , K t , ℓ t

∂ n(Zt , X t )
 I II
ˆ ˆ
<0 if X t ∈ X t , X t
∂ Xt

ˆ II
For X t > X t , the q q t locus is horizontal at q t = 0 since net fertility nt is equal to unity.
 I II
ˆ ˆ
For X t ∈ X t , X t , the q q t locus is downward-sloping with decreasing n(Zt , X t ) and
ˆI
increasing f N M (Zt , X t ) with respect to X t . For X t < X t , however, the shape of the q q t
locus remains ambiguous because n(Zt , X t ) is increasing in X t in this case. The sign of

104
 qt

q tmax

q qt

0 Xt
0 ˆ I
ˆ II
Xt Xt

Figure B.2: The q q t Locus and the Dynamics of q t

the slope of the q q t locus is equal to the sign of

€ Š € Š
nX (Zt , X t ) f N M (Zt , X t ) − ρ/st ℓ t − fXN M (Zt , X t ) n(Zt , X t ) − 1 ⋛ 0

where nX (•, •) and fXN M (•, •) denote associated partial derivatives with respect to X t . An
inspection of second derivatives with respect to X t further indicates that there may exist
ˆI
local maxima and local minima of q t for 0 < X t < X t .

The ambiguity, fortunately, does not affect the global stability result. The reason, as
ˆI
it shall become clear below, is that, for any 0 < X t < X t , we have

– ™
φℓ t st
n(Zt , X t ) ∈ 1, and f N M (Zt , X t ) ∈ (0, 1]
ρ(φ + γ ) + ψℓ t

These imply, together with the uniqueness of q t given (Zt , X t ), that there exists a unique
ˆI
global maximum q tmax < ∞ of the q q t locus for 0 < X t < X t regardless of its local

105
 qt

X Xt

q̂ t

q tmax

q qt

0 Xt
0 ˆ I
ˆ II
Xt Xt

Figure B.3: The Convergence to the Asymptotic Equilibrium

maxima and local minima.

Figure B.2 pictures one possible characterization of the q q t locus and the associated
dynamics of q t . With the confidence following from the existence of unique q tmax which
is bounded above, the rest of the analysis is carried out with this possibility.

In Figure B.3, the X X t set and the q q t locus are drawn together. Note that the fol-
lowing are subject to change in time due to the dependence on Zt :

• the X X t set,

• the q̂ t threshold,

• the q q t locus,

• the q tmax maximum,

106
 qt

X Xt
q tmax A

q̂ t

q qt

0 Xt
0 ˆI
ˆ II
Xt Xt

Figure B.4: A Quasi-Static Equilibrium without Productivity Growth

ˆI ˆ II
• the X t and X t thresholds.

The claim here is that, even though these elements of the conditional dynamical sys-
tem change in not easily predictable ways because of evolving Zt , q̂ t becomes strictly
greater than q tmax , once and for all, at some finite t . To see why, notice that net fertility
ˆ II
nt is greater than unity for any X t that is less than X t . This in turn implies that q̂ t

grows, even for constant ℓ t , as long as X t is sufficiently low. Put differently, the set X X t
continuously moves upwards on (X t , q t ) plane along the trajectory towards the industrial
revolution.38 Once the system is characterized by q̂ t > q tmax , it always moves towards the
asymptotic equilibrium of X t → ∞ and q t → 0 because the qualitative properties of the

system remains unchanged. When X t passes its second threshold, nt = n ⋆ = 1 lets q̂ t

stabilize through N t = N ⋆ and ℓ t converging to unity.

In Figure B.4, a quasi-static equilibrium without productivity growth in the mod-

ern sector is pictured to provide more insight on the dynamic properties of the model.

38. Needless to say, assuming that the industrial revolution is possible implies q̂ t > 0 for all t .

107
Recalling the earlier characterization of the initial stagnation equilibrium, if it is again
assumed that st and ℓ t are constant and that the exogenous productivity growth in the
traditional sector just matches the effect of slowly increasing population, the point A in
the figure represents a quasi-static equilibrium of the model. In this equilibrium, a knife-

edge relation is endogenously established between population growth and the growth of
the stock of useful discoveries, making q t +1 = q t . Since the labor share of the traditional
sector does not change in this equilibrium, the quasi-statis can be prolonged in calendar
time until the stock of useful discoveries, or, equivalently, the level of population given
constant q t , is sufficiently large to imply q̂ t > q tmax .

108
 References

Acemoglu, D. and F. Zilibotti (1997), “Was Prometheus Unbound by Chance? Risk,

Diversification, and Growth,” Journal of Political Economy 105(4), 709–751.

Aghion, P. and P. Howitt (1992), “A Model of Growth through Creative Destruction,”

Econometrica 60(2), 323–351.

Aghion, P. and P. Howitt (1998), Endogenous Growth Theory Cambridge: MIT Press.

Aghion, P. and P. Howitt (2009), The Economics of Growth Cambridge: MIT Press.

Akcigit, U. and W. R. Kerr (2010), “Growth Through Heterogeneous Innovations,”

NBER Working Paper No. 16443.

Allen, R. C. (2011), “Why the industrial revolution was British: commerce, induced
invention, and the scientific revolution,” Economic History Review 64(2), 357–384.

Arifovic, J., J. Bullard, and J. Duffy (1997), “The Transition from Stagnation to Growth:
An Adaptive Learning Approach,” Journal of Economic Growth 2(2), 185–209.

Azariadis, C. and A. Drazen (1990), “Threshold Externalities in Economic Develop-

ment,” Quarterly Journal of Economics 105(2), 501–526.

Bar, M. and O. Leukhina (2010aa), “Demographic transition and industrial revolution:

A macroeconomic investigation,” Review of Economic Dynamics 13(2), 424–451.

Bar, M. and O. Leukhina (2010bb), “The role of mortality in the transmission of knowl-
edge,” Journal of Economic Growth 15(4), 291–321.

Baumol, W. J. (1990), “Entrepreneurship: Productive, Unproductive, and Destructive,”

Journal of Political Economy 98(5), 893–921.

Becker, G. S. and R. J. Barro (1988), “A Reformulation of the Economic Theory of Fer-

tility,” The Quarterly Journal of Economics 103(1), 1–25.

Becker, G. S., K. M. Murphy, and R. Tamura (1990), “Human Capital, Fertility, and
Economic Growth,” Journal of Political Economy 98(5), S12–S37.

Bekar, C. T. and R. G. Lipsey (2004), “Science, Institutions and the Industrial Revolu-
tion,” Journal of European Economic History 33(3), 709–753.

Boldrin, M. and L. E. Jones (2002), “Mortality, Fertility, and Saving in a Malthusian

Economy,” Review of Economic Dynamics 5(4), 775–814.

Boserup, E. (1965), The Conditions of Agricultural Progress Chicago: Aldine Publishing

Company.

109
Cervellati, M. and U. Sunde (2005), “Human Capital Formation, Life Expectancy, and
the Process of Development,” American Economic Review 95(5), 1653–1672.
Clark, G. (2001), “The Secret History of the Industrial Revolution,” Working Paper.
Clark, G. (2002), “The Agricultural Revolution and the Industrial Revolution: England,
1500-1912,” Working Paper.
Clark, G. (2005), “The Condition of the Working Class in England, 1209-2004,” Journal
of Political Economy 113(6), 1307–1340.
Clark, G. (2007), A Farewell to Alms: A Brief Economic History of the World Princeton:
Princeton University Press.
Clark, G. (2009), “The Macroeconomic Aggregates for England, 1209-2008,” Working
Paper.
Crafts, N. F. R. (1977), “Industrial revolution in England and France: Some thoughts on
the question ’Why Was England First?’,” Economic History Review, Second Series 30(3),
429–441.
Crafts, N. F. R. (2005), “The First Industrial Revolution: Resolving the Slow
Growth/Rapid Industrialization Paradox,” Journal of the European Economic Associ-
ation 3(2-3), 525–534.
Crafts, N. F. R. and C. K. Harley (1992), “Output growth and the British Industrial
Revolution: a restatement of the Crafts-Harley view,” Economic History Review 45(4),
703–730.
Crafts, N. F. R. and C. K. Harley (2000), “Simulating the two views of the British Indus-
trial Revolution,” Journal of Economic History 60(3), 819–842.
Crafts, N. F. R. and T. C. Mills (2009), “From Malthus to Solow: How did the Malthusian
economy really evolve?,” Journal of Macroeconomics 31(1), 68–93.
de la Croix, D. and O. Licandro (2009), “The Child is Father of the Man: Implications
for the Demographic Transition,” Working Paper.
de la Croix, D. and A. Sommacal (2009), “A Theory of Medical Effectiveness, Differential
Mortality, Income Inequality and Growth for Pre-Industrial England,” Mathematical
Population Studies 16(1), 2–35.
Desmet, K. and S. Parente (2009), “The Evolution of Markets and the Revolution of
Industry: A Quantitative Model of England’s Development, 1300-2000,” Working
Paper.
Dinopoulos, E. and P. Thompson (1998), “Schumpeterian Growth without Scale Effects,”
Journal of Economic Growth 3(4), 313–335.
Doepke, M. (2004), “Accounting for Fertility Decline During the Transition to Growth,”
Journal of Economic Growth 9(3), 347–383.

110
Doepke, M. and F. Zilibotti (2008), “Occupational Choice and the Spirit of Capitalism,”
Quarterly Journal of Economics 123(2), 747–793.
Easterlin, R. A. (1995), “Industrial Revolution and Mortality Revolution: Two of a
Kind?,” Journal of Evolutionary Economics 5(4), 393–408.
Galor, O. (2005), “From Stagnation to Growth: Unified Growth Theory,” In: P. Aghion
and S. N. Durlauf (eds.), Handbook of Economic Growth, Vol. 1A Amsterdam: Elsevier,
Ch. 4, pp. 171–293.
Galor, O. (2010), “The 2008 Lawrence R. Klein Lecture-Comparative Economic Devel-
opment: Insights From Unified Growth Theory,” International Economic Review
51(1), 1–44.
Galor, O. and S. Michalopoulos (2009), “The Evolution of Entrepreneurial Spirit and the
Process of Development,” Working Paper.
Galor, O. and O. Moav (2002), “Natural Selection And The Origin Of Economic
Growth,” Quarterly Journal of Economics 117(4), 1133–1191.
Galor, O. and D. N. Weil (2000), “Population, Technology, and Growth: From Malthu-
sian Stagnation to the Demographic Transition and Beyond,” American Economic Re-
view 90(4), 806–828.
Goodfriend, M. and J. McDermott (1995), “Early Development,” American Economic
Review 85(1), 116–133.
Grossmann, V. (2009), “Entrepreneurial innovation and economic growth,” Journal of
Macroeconomics 31(4), 602–613.
Hansen, G. D. and E. C. Prescott (2002), “Malthus to Solow,” American Economic Review
92(4), 1205–1217.
Hanson, R. (1998), “Patterns of Patronage: Why Grants Won Over Prizes in Science,”
Working Paper.
Haruyama, T. (2009), “Competitive Innovation With Codified and Tacit Knowledge,”
Scottish Journal of Political Economy 56(4), 390–414.
Hazan, M. and B. Berdugo (2002), “Child Labour, Fertility, and Economic Growth,”
Economic Journal 112(482), 810–828.
Hazan, M. and H. Zoabi (2006), “Does longevity cause growth? A theoretical critique,”
Journal of Economic Growth 11(4), 363–376.
Hellwig, M. and A. Irmen (2001), “Endogenous Technical Change in a Competitive
Economy,” Journal of Economic Theory 101(1), 1–39.
Howitt, P. and D. Mayer-Foulkes (2005), “R&D, Implementation, and Stagnation: A
Schumpeterian Theory of Convergence Clubs,” Journal of Money, Credit and Banking
37(1), 147–177.

111
Jacob, M. C. (1997), Scientific Culture and the Making of the Industrial West New York:
Oxford University Press.

Jones, C. I. (2001), “Was an Industrial Revolution Inevitable? Economic Growth Over

the Very Long Run,” Advances in Macroeconomics 1(2) Article 1.

Jones, E. L. (2010), Locating the Industrial Revolution: Inducement and Response London:
World Scientific Publishing Company.

Jones, L. E., A. Schoonbroodt, and M. Tertilt (2011), “Fertility Theories: Can They Ex-
plain the Negative Fertility-Income Relationship?,” In: J. B. Shoven (ed.), Demography
and the Economy, Chicago: University of Chicago Press, Ch. 2, pp. 43–100.

Kelly, M. (2005), “The Invention of Invention,” Working Paper.

Klette, T. J. and S. Kortum (2004), “Innovating Firms and Aggregate Innovation,” Journal
of Political Economy 112(5), 986–1018.

Kögel, T. and A. Prskawetz (2001), “Agricultural Productivity Growth and Escape from
the Malthusian Trap,” Journal of Economic Growth 6(4), 337–357.

Kremer, M. (1993), “Population Growth and Technological Change: One Million B.C.
to 1990,” Quarterly Journal of Economics 108(3), 681–716.

Lagerlöf, N.-P. (2003), “From Malthus to Modern Growth: Can Epidemics Explain the
Three Regimes?,” International Economic Review 44(2), 755–777.

Landes, D. S. (1969), The Unbound Prometheus: Technological Change and Industrial De-
velopment in Western Europe from 1750 to the Present Cambridge: Cambridge Univer-
sity Press.

Landes, D. S. (2006), “Why Europe and the West? Why Not China?,” Journal of Economic
Perspectives 20(2), 3–22.

Lazonick, W. (1991), “What Happened to the Theory of Economic Development?,” In:

P. Higgonet, D. S. Landes, and H. Rosovsky (eds.), Favorites of Fortune: Technology,
Growth, and Economic Development since the Industrial Revolution Cambridge: Har-
vard University Press, Ch. 9, pp. 267–296.

Lipsey, R. G., K. I. Carlaw, and C. T. Bekar (2005), Economic Transformations: General

Purpose Technologies and Long Term Economic Growth New York: Oxford University
Press.

Lucas, Jr., R. E. (1978), “On the Size Distribution of Business Firms,” Bell Journal of
Economics 9(2), 508–523.

Lucas, Jr., R. E. (2002), “The Industrial Revolution: Past and Future,” In: Lectures on
Economic Growth, Cambridge: Harvard University Press, Ch. 5, pp. 97–188.

Lucas, Jr., R. E. (2009), “Ideas and Growth,” Economica 76(301), 1–19.

112
Maddison, A. (1995), Monitoring the World Economy, 1820-1992 Paris: OECD Develop-
ment Centre.

Madsen, J., J. Ang, and R. Banerjee (2010), “Four centuries of British economic growth:
the roles of technology and population,” Journal of Economic Growth 15(4), 263–290.

Matsuyama, K. (1991), “Increasing Returns, Industrialization, and Indeterminacy of

Equilibrium,” Quarterly Journal of Economics 106(2), 617–50.

Meisenzahl, R. R. and J. Mokyr (forthcoming), “The Rate and Direction of Invention

in the British Industrial Revolution: Incentives and Institutions,” In: J. Lerner and S.
Stern (eds.), Rate and Direction of Inventive Activity Chicago: University of Chicago
Press.

Michelacci, C. (2003), “Low Returns in R&D Due to the Lack of Entrepreneurial Skills,”
Economic Journal 113(484), 207–225.

Milionis, P. and M. Klasing (2009), “Innovation-Based Growth & Long-Run Economic

Development,” Working Paper.

Mitchell, B. R. (1975), European Historical Statistics 1750-1970 New York: Columbia

University Press.

Mokyr, J. (1993), “Technological Progress and the Decline of European Mortality,” Amer-
ican Economic Review 83(2), 324–330.

Mokyr, J. (1998), “Editor’s introduction: The new economic history and the Industrial
Revolution,” In: J. Mokyr (ed.), The British Industrial Revolution: An Economic Per-
spective Boulder: Westview Press, pp. 1–127.

Mokyr, J. (2002), The Gifts of Athena: Historical Origins of the Knowledge Economy Prince-
ton: Princeton University Press.

Mokyr, J. (2004), “Accounting for the Industrial Revolution,” In: R. Floud and P. John-
son (eds.), The Cambridge Economic History of Modern Britain, Vol. 1 Cambridge:
Cambridge University Press, Ch. 1, pp. 1–27.

Mokyr, J. (2010), “Entrepreneurship and the Industrial Revolution in Britain,” In: D. S.

Landes, J. Mokyr, and W. J. Baumol (eds.), The Invention of Enterprise: Entrepreneur-
ship from Ancient Mesopotamia to Modern Times Princeton: Princeton University Press,
Ch. 7, pp. 183–210.

Murphy, K. M., A. Shleifer, and R. W. Vishny (1989), “Industrialization and the Big
Push,” Journal of Political Economy 97(5), 1003–1026.

Murphy, K. M., A. Shleifer, and R. W. Vishny (1991), “The Allocation of Talent: Impli-
cations for Growth,” Quarterly Journal of Economics 106(2), 503–530.

Nelson, R. R. (1982), “The Role of Knowledge in R&D Efficiency,” Quarterly Journal of

Economics 97(3), 453–470.

113
O’Rourke, K. H., A. S. Rahman, and A. M. Taylor (2008), “Luddites and the Demo-
graphic Transition,” NBER Working Paper No. 14484.

Pereira, A. S. (2003), “Essays on the Origins of Modern Economic Growth,” Ph.D. thesis,
Simon Fraser University.

Peretto, P. F. (1998aa), “Technological Change and Population Growth,” Journal of Eco-

nomic Growth 3(4), 283–311.

Peretto, P. F. (1998bb), “Technological Change, Market Rivalry, and the Evolution of the
Capitalist Engine of Growth,” Journal of Economic Growth 3(1), 53–80.

Peretto, P. F. and S. Valente (2011), “Growth on a Finite Planet: Resources, Technology

and Population in the Long Run,” Working Paper.

Pomeranz, K. (2000), The Great Divergence: China, Europe, and the Making of the Modern
World Economy Princeton: Princeton University Press.

Pumfrey, S. and F. Dawbarn (2004), “Science and Patronage in England, 1570-1625: A

Preliminary Study,” History of Science 42(136), 137–188.

Rosenberg, N. (1974), “Science, Invention and Economic Growth,” Economic Journal

84(333), 90–108.

Routh, G. (1987), Occupations of the People of Great Britain 1801-1981 London: The
Macmillan Press Ltd.

Schumpeter, J. A. (1934), The Theory of Economic Development Cambridge: Harvard

University Press (First German Edition in 1912.).

Solo, C. S. (1951), “Innovation in the Capitalist Process: A Critique of the Schumpeterian

Theory,” Quarterly Journal of Economics 65(3), 417–428.

Strulik, H. (2009), “Knowledge and Growth in the Very Long-Run,” Working Paper.

Strulik, H. and J. Weisdorf (2008), “Population, food, and knowledge: a simple unified
growth theory,” Journal of Economic Growth 13(3), 195–216.

Strulik, H. and J. Weisdorf (2011), “How Child Costs and Survival Shaped the Industrial
Revolution and the Demographic Transition: A Theoretical Inquiry,” Working Paper.

Sullivan, R. J. (1989), “England’s Age of invention: The acceleration of patents and

patentable invention during the industrial revolution,” Explorations in Economic His-
tory 26(4), 424–452.

Tamura, R. (1996), “From decay to growth: A demographic transition to economic

growth,” Journal of Economic Dynamics and Control 20(6-7), 1237–1261.

Tamura, R. (2002), “Human capital and the switch from agriculture to industry,” Journal
of Economic Dynamics and Control 27(2), 207–242.

114
Taylor, P. J., M. Hoyler, and D. M. Evans (2008), “A Geohistorical Study of ’The Rise of
Modern Science’: Mapping Scientific Practice Through Urban Networks, 1500-1900,”
Minerva 46(4), 391–410.

Temin, P. (1997), “Two Views of the British Industrial Revolution,” Journal of Economic
History 57(1), 63–82.

Voigtländer, N. and H.-J. Voth (2006), “Why England? Demographic factors, structural
change and physical capital accumulation during the Industrial Revolution,” Journal
of Economic Growth 11(4), 319–361.

Voth, H.-J. (2003), “Living Standards During the Industrial Revolution: An Economist’s
Guide,” American Economic Review 93(2), 221–226.

Weitzman, M. L. (1998), “Recombinant Growth,” Quarterly Journal of Economics 113(2),

331–360.

Wrigley, E. A., R. S. Davies, J. E. J. E. Oeppen, and R. S. Schofield (1997), English Popula-

tion History from Family Reconstitution 1580-1837 Cambridge: Cambridge University
Press.

Young, A. (1998), “Growth without Scale Effects,” Journal of Political Economy 106(1),
41–63.

115