The
Econometricians
Gauss, Galton, Pearson, Fisher, Hotelling, Cowles,
Frisch, and Haavelmo
Great Minds in Finance
Series Editor
Professor ColinRead
Professor of Economics and Finance
former Dean of the School of Business and Economics
The State University of New York
at Plattsburgh (SUNY), USA
Aims of the Series
This series explores the lives and times, theories and applications of those
who have contributed most significantly to the formal study of finance.
It aims to bring to life the theories that are the foundation of modern
finance, by examining them within the context of the historical backdrop
and the life stories and characters of the 'great minds' behind them.
Readers may be those interested in the fundamental underpinnings of
our stock and bond markets; college students who want to delve into the
significance behind the theories; or experts who constantly look for ways
to more clearly understand what they do, so they can better relate to their
clients and communities.
The Econometricians
Gauss, Galton, Pearson, Fisher, Hotelling, Cowles,
Frisch and Haavelmo
ColinRead
Professor of Economics and Finance
former Dean of the School of Business and Economics,
The State University of New York at Plattsburgh (SUNY), USA
This book is the seventh in a series of discussions about the great minds
in the history and theory of finance.
The series describes the contributions of those remarkable individuals
who expanded our understanding of the underpinnings and theories of
modern finance. While earlier volumes discussed those who described
the importance of growth and interest rates on our economic decisions,
described the methods by which we choose assets for our portfolios, dis-
cussed whether markets are efficient, and described the roots and applica-
tions of public finance, this volume treats the statistical and econometric
tools that provide the foundation of all finance theory.
The statisticians and econometricians who we describe here collec-
tively created the framework and the techniques with which we have
viewed finance ever since. Their tools were, at times, developed to solve
problems that were intuitively unrelated to finance, but which were even-
tually tailored to the unique needs of the discipline. A series of great
minds developed their concepts well before theorists and practitioners
had access to modern computing, so their theories were necessarily intui-
tive and geometric, and easy to apply with a pencil and paper and some
incredible imagination. While the technological limitations of their day
had initially limited and somewhat rigidly defined how we now view our
financial variables and analyses, the limitations also offer a simplicity and
create a vocabulary that is most accessible.
v
vi Preamble
conclusions that offer little utility. On the contrary, the best theories
draw the strongest possible conclusions from the weakest set of assump-
tions. And, a successful unifying theory in finance can replace a large
number of lesser theories and concepts, just as physicists hold out for
a unifying theory that can draw together their isolated understandings
from a variety of specialties.
By focusing on the great minds in finance, we draw together the con-
cepts that have stood the test of time and have proven themselves to
reveal something about the way humans make financial decisions. These
principles that have flowed from individuals who are typically awarded
the Nobel Memorial Prize in Economics for their insights, or perhaps
shall be awarded someday, allow us to see the financial forest for the trees.
While one might assume that every financial expert would be well
versed in these fundamentals, such is not the case. An investor can suc-
ceed through sheer intuition without having studied the insights of theo-
rists over a century of financial discovery. Mathematicians and physicists
are increasingly employed to develop techniques that recognize patterns
in numbers with little regard or understanding of the underlying forces
that explain these patterns. And, computer experts can design algorithms
that allow great banks of servers to constantly poke and prod the market
to induce, and then profit from, movements in prices of stocks or bonds.
By capitalizing on such shifts in prices milliseconds before others take
notice, these algorithms can garner pennies, or fractions of pennies, at a
time, thousands of times an hour, to yield huge profits.
These practitioners do not depend on, or even care about, the funda-
mental principles that drive markets in the long run. To them, the long
run expires within a week or a day. Such technical analysis is decidedly
transient and short term. In fact, a steady and predictable investment
opportunity based on well-known and well-understood information is
simply insufficiently volatile to yield quick profits.
Unfortunately, such technical analysis that depends only on price
dynamics in the short term has emerged as the lucrative Holy Grail of
modern finance. It allows the most skilled practitioners to make money
when markets are rising or falling. However, it reveals nothing about
how financial decisions should be made in the long run to satisfy an
economys need for capital, investment, reward and reduced risk.
xii Preface to the Great Minds in Finance Series
Nor does it make our economy more efficient. Rather, technical analysts
devote a great deal of talent, energy and effort as they clamor for others
pieces of a fixed economic pie.
The giants who have produced the theories and concepts that drive
financial fundamentals share one important characteristic. They have
developed insights that explain how markets can be used or tailored to
create a more efficient economy. They demonstrate how individuals can
trade risk and reward in the same way that a supplier might trade with a
consumer of a good. Through this process, both sides win. Greater effi-
ciency is a tide that lifts all boats. These pioneers of finance explain how
tools can be used to create greater market efficiency and even suggest the
creation of new tools to create efficiency enhancements that may have
proven elusive otherwise.
Global financial markets are experiencing a technological revolution.
From a strictly aesthetic perspective, one cannot entirely condemn the
tug-of-war struggle for profits the technicians seek, even if they do little
to enhance, and may even detract from, efficiency. The mathematics and
physics of price movements and the sophistication of computer algo-
rithms are fascinating in their own rights. Indeed, my university studies
began with a Bachelor of Science degree in physics, followed by a PhD in
economics. However, as I began to teach economics and finance, I real-
ized that the analytic tools of physics that so pervades theories of modern
finance has strayed too far from explaining the essence of human finan-
cial decision-making.
As I taught the economics of intertemporal choice, the role of money
and financial instruments, and the structure of the banking and financial
intermediaries, I also recognized that my students had become increas-
ingly fascinated with investment banking and Wall Street. Meanwhile,
the developed world experienced the most significant breakdown of
financial markets in almost eight decades. I realized that this once-in-
a-lifetime global financial meltdown arose because we had moved from
an economy that produced stuff to one in which a third of all profits by
2006in the USA were made in the financial industry, with little to show
but pieces of paper representing wealth that had value only if some were
ever ready to buy them.
Preface to the Great Minds in Finance Series xiii
xv
xvi Contents
31 Conclusions 247
Glossary 251
Index 257
About the Author
xix
List of Figures
xxi
Part 1
Mathematicians and Astronomers
Cathrine Heuer
Carl Friedrich Gauss
b:
b: 30 Apr 1777 in Braunschweig,
Braunschweig, Niedersachsen, d: ; Y
Germany
m: 09 Oct 1805 in Brunswick,
Braunschweig, Niedersachsen, Christoph Benze Andreas Bentze Andreas Bentze
Germany
d: 23 Feb 1855 in , Gottingen, b: 02 Mar 1717 in Velpke, Germany b: 04 Feb 1687 in Velpke, Germany
Niedersachsen, Germany m: m: 15 Jan 1715 in Velpke, Germany
d: 01 Sep 1748 in Velpke, d: 02 Mar 1730 in Velpke, Germany
Helmstedt, Niedersachsen,
Germany Marie Elizabeth Suepke
Dorthea Benze b: 20 Apr 1693 in Velpke, Germany
b: 18 Jun 1743 in Velpke, Germany d: 02 Nov 1729 in Velpke, Germany
d: 18 Apr 1839 in Gottingen,
Germany
Katharina Maria Crone
b: 1710
d: 1771
to guilds, and all would provide him with employment through the four
seasons. The family enjoyed a level of urban economic comfort not read-
ily afforded to the youngest child in a farming family.
Jurgen Gauss and Katharine soon secured a small and narrow house
for his family, at 10 Ritterbrunnen. They lived in that home for 14 years,
and raised four children there, including an eldest son, Gebhard Dietrich
Gauss (13 February 174414 April 1808). The family then moved to a
larger home at 30 Wilhelmstrasse, where Jurgen would die of tuberculosis
on 5 July 1774, just three months and two days after the death of his wife
from a prolonged fever.
By the time his father died, the second child and the eldest son,
Gebhard, had worked and learned the family trades. Upon his parents
death, Gebhard used a dowry from his first wife, Dorothea Emerenzia
Warnecke, and a loan from the towns mayor Wilmerding, to buy out the
shares of his family home from his brothers Johanne Franz Heinrich and
Peter Heinrich. By the age of 30, Gebhard was able to provide a home
and a secure but not affluent living for his own family.
6 The Econometricians
Gebhards first wife did not long enjoy the house, though. She died on
5 September 1775 of tuberculosis as had her father-in-law, but not before
giving birth to a boy. Johann Georg Heinrich was born on 14 January
1769.
Seven months after Dorothea Warnecke died, Gebhard married
Dorothea Benze, the daughter of a stonemason, Christophe Benze (2
March 17171 September 1748). Dorotheas father died prematurely as
well, from pulmonary respiratory illness associated with his profession as
a stonemason. He also left a son, Johann Friedrich. Both of Christophe
Benzes children were thoughtful and intelligent, but neither enjoyed the
luxury of formal schooling.
Dorothea was illiterate, but she was a kind and nurturing woman
by nature. She worked as a maid before she married Gebhard. While
Gebhard was also uneducated, he nonetheless managed to be appointed
the citys master of waterworks as he was an experienced stonemason and
was reasonably good with sums.
In contrast to his wifes gentle nature, Gebhard was quite domineering
as a father, and was considered rather somewhat uncouth. Yet, he pro-
vided reasonably well for his family.
Carl had great affection for his mother, and for her brother, his uncle
Johann Friedrich, but had a somewhat awkward relationship with his
half brother. Indeed, he did not know his brother Johann Georg very
well. More than eight years his senior, Johann struck out on his own as
a day laborer before Carls tenth birthday. Johann returned home some
years later, but an eye injury made him only of limited help to his father.
Instead, Johann enlisted in the army for almost a decade, and returned
to his family home to take over his fathers trade once his father died on
14 April 1808.
While Carl Friedrich grew up with a harsh and domineering father,
he enjoyed the better nature of his kind and devoted mother. In turn, he
doted on his mother all his life, until Dorotheas death at the remarkably
advanced age of 97, even through her infirmities and her affliction with
blindness in her last four years.
Carl harbored fond and vivid memories of his childhood. One of his
earliest memories was falling into the river and being saved at a very
young age. This terrifying memory did not taint his recollections other-
wise, though.
He also recalled that he taught himself to read by asking his family
members how to pronounce letters on the page. His ability to manipu-
late numbers became a favorite parlor trick among family friends. When
his father would pay his bricklaying workers, a three-year-old Carl once
impudently but accurately corrected his fathers calculations. While
access to school was out of the question for many children in his neigh-
borhood, his intellectual precociousness compelled his family to consider
his education.
In 1784, when young Carl was seven years of age, his father agreed to
let him enter nearby St. Katharines Volksschule, a peoples school for the
more motivated of the children of non-prosperous families. The school,
which adjoined St. Katharines Church, had a teacher named J.G.Buttner
who oversaw a dark and dank classroom of 200 children.
Young Gauss endured two years among those 200 classmates, but
managed to stand out nonetheless. When the teacher gave his students
an assignment he felt would keep them occupied for hours, Gauss almost
immediately announced the solution. The teacher felt it impossible that
8 The Econometricians
Gauss could have so quickly added all the numbers from 1 to 100. Yet,
an eight-year-old Gauss reasoned his way to a solution.
Gauss had noticed that the first and last number in the sequence, 1
and 100, added to 101. So did the second and the second to last, 2 and
99, and, indeed, so do all 50 such pairs. Gauss proclaimed that 50 pairs
which each add each to 101 must then total 5050. While he may not
have been praised for his brilliance, he was at least spared the whip that
often accompanied the incorrect answers from lesser classmates.
Headmaster Buttner quickly realized Gauss needed a more sophisti-
cated curriculum. He ordered an advanced textbook for the boy, but pro-
fessed that there was little he could teach the young prodigy. Fortunately,
word got out in Brunswick about the boy wonder. Carls father relented
on the headmasters admonishments that Carl spend the evening study-
ing rather than completing chores, and allowed the neighbors boy,
Johann Christian Martin Bartels (17691836), eight years Carls senior,
and with a strong interest and competency in mathematics, to supple-
ment Carls education. By lamplight, Carl learned mathematics together
with the much older boy, and kept pace as Bartels absorbed such concepts
as the binomial theorem and infinite series.
Meanwhile, Bartels worked to find a patron for Carl as he pursued his
own studies in mathematics. His Collegium teacher, Eberhard August
Wilhelm Zimmermann (17 August 17434 July 1815), had studied
mathematics at the prestigious Gttingen University and was instructing
mathematics at the Collegium Carolinum in Brunswick. When Bartels
began classes with Zimmerman, he brought the abilities of the 11-year-
old Carl Gauss to his professors attention. By then, Zimmermann was
well respected by the local nobility, and had already been awarded the
distinction of Councilor in the region. Seven years later, he was further
bestowed a noble title by Duke Karl hette Wilhelm Ferdinand.
Zimmerman accepted the challenge of instructing the young prodigy.
Carl Gauss was given the run of the Collegium, despite his young age,
and spent almost all his free time at the school.
A few years later, the Duchess Ferdinand came across the young boy
reading a book that seemed most advanced for his age. Astonished, she
quizzed him on his studies, and was impressed by his knowledge of the
classics, of literature and of mathematics.
1 The Early Life ofCarl Friedrich Gauss 9
At the Duchess behest, the Duke sent an aide to fetch young Gauss.
The aide first demanded that Carls older brother accompany him to the
palace, but Johann Georg insisted it was for his young half brother Carl
for whom the Duke beckoned.
The Duke took under his wing a young working-class boy, one with
rough working-class edges who spoke Low German. Despite these social
handicaps, the Duke funded Carls regular attendance at the Collegium
Carolinum in 1792. Only 15 years old, Carl was already more intellectu-
ally advanced than most of the other students. In addition to the annual
stipend the Duke paid on behalf of Carl, the Duke also offered Carls
teacher Zimmerman expenses to oversee his education, for as long as
Gauss attended the local college.
Young Carl studied at the Collegium Carolinum for four years. He was
exposed to the Classics, to Greek and Latin and to the mathematics of Sir
Isaac Newton (25 December 164220 March 1726), Leonard Euler (15
April 170718 September 1783) and Joseph-Louis Lagrange (25 January
173610 April 1813). By his last year at the college, Gauss had become
intellectually captivated by the astronomical explorations of Newton,
and had even developed his method of least squares to tease trends from
cosmological observations subject to random errors. While also still a
teenager at the College, Gauss also became intrigued with prime num-
bers, a fascination he would continue soon upon his matriculation to the
University of Gttingen.
It was at the Collegium that Carl also developed the habit of intellec-
tual innovation primarily for his own curiositys sake. He only occasion-
ally felt the need to aggregate his ideas into papers, even though he soon
realized the need to record his results. He meticulously began to docu-
ment his mathematical innovations in a series of dated notebook entries
that he kept for his entire life.
On 21 August 1795, when Carl was 18 years old, the Duke ordered
an increase in the stipend to be paid to Carl so that Carl could begin
attending at the University. This sum covered tuition and all living
expenses. On 11 October, Gauss left his hometown for the 200-kilome-
ter trip to Gttingen, one of the nations most prestigious and accom-
plished universities. Four days after departing Brunswick, Gauss arrived
and was admitted as a mathematics student to Gttingen. Carl chose the
10 The Econometricians
placed its host city at the center of learning in Germany ever since. The
brilliant mathematicians Bernhard Riemann (17 September 182620
July 1866), David Hilbert (23 January 186214 February 1943), Peter
Gustav Lejeune Dirichlet (13 February 18055 May 1859) and John von
Neumann (28 December 19038 February 1957), and great physicists
such as Max Born (11 December 18825 January 1970), Julius Robert
Oppenheimer (22 April 190418 February 1967), Max Planck (23 April
18584 October 1947), Werner Karl Heisenberg (5 December 19011
February 1976), Enrico Fermi (29 September 190128 November 1954)
and Wolfgang Pauli (25 April 190015 December 1958) all studied or
taught there. So did the international banker and financier John Pierpont
J.P. Morgan (17 April 183731 March 1913). It was also the notorious
epicenter of Adolf Hitlers (20 April 188930 April 1945) Great Purge of
Jewish academics in 1933. Gttingen was an intellectual capital unpar-
alleled in the celebration of abstract thought since the early 1800s, or
perhaps since Carl Friedrich Gauss first studied there.
Those accepted to study at Gttingen were invariably gifted. But, few
came from such modest means as had Gauss, nor with such wealthy
patrons as Duke Ferdinand. When Gauss was admitted at Gttingen on
15 October 1795,1 at the age of 18, he did not yet know whether he
wished to study mathematics or philology. Gauss loved words almost as
much as he knew numbers.
Philology was a classical and well-appreciated discipline in the nine-
teenth century that divined language from the historical written record.
Philologists were Renaissance scholars who used their skills in history,
linguistics and literary criticism to solve literary and historical puzzles. It
was at that time a foundation for what might more broadly be described
as the humanities today. The disciplines goal to solve historical and liter-
ary puzzles played to Gauss curiosity in the same way as numbers did,
and drew upon Gauss extensive knowledge of many languages as had his
ability to draw upon many methodologies in mathematics.
Fortunately for modern science, mathematics won Gauss attention,
partly because the patronage he enjoyed allowed him to be less concerned
about tuition and eventual salaries, and more receptive to the study of
science, with all its economic impracticality. During his first year of
study, he mostly read books from the renowned Gttingen library on
2 The Times ofCarl Friedrich Gauss 13
the humanities, on philology and on travel. But, that first year sparked a
passion in mathematics when Gauss managed to devise his first solution
to a then intractable mathematical problem. He had become fascinated
in the problem of dividing a circle into 17 parts through the creation of a
17-sided polygon constructed only with Euclidean tools.
It is instructive to explore Gauss path through the construction of his
problem and solution as it tells us much about the brilliance of this 18
year old, on the development of the new field of complex algebra ever
since, and on the way Gauss thought, in geometric terms that would
become the hallmark of the greatest minds of statistics and econometrics.
The Greeks before the birth of Jesus Christ had been fascinated with
the geometries that could be constructed with a simple compass and
straightedge. Indeed, these early geometers had no number system yet.
These tools of the compass and straightedge were what Euclid (about
300 BC) employed to construct regular polygons. These regular polygons
have angles between apexes that are all equal, and sides that are the same
length, such as triangles, squares, hexagons and octagons.
The Greeks were fascinated by the properties of such polygons that
were contained within a circle. By calculating the area of regular polygons
of ever increasing number of sides, they could even approximate the area
of a circle and the number pi with great accuracy.
The Greeks quickly discovered that they could easily construct such
polygons with an even number of sides by forming isosceles triangles with
two equal sides in the space from the center of the circle to its circumfer-
ence. They could easily construct even-number-sided polygons within the
unit circle by further subdividing known polygons with an even number
of sides. Such exercises allowed the Greeks to construct extensive proofs
of the properties of lines, circles, triangles, squares and octagons.
While they did not actually develop a number system from the length
of the sides of these polygons, they were clearly dabbling in number the-
ory. We are reminded of this geometric interpretation when we think of
raising a number to the power of two as squaring the number. We can
now see these Greek geometers were on the verge of discovering alge-
bra, polynomials, roots of polynomials, negative numbers and imaginary
numbers. But, that leap in understanding would take almost two millen-
nia to solve.
14 The Econometricians
For instance, consider the area of a square drawn from the apex F of a
semicircle of width AX and height AF=AX, and hence of area AX*AX.
The Greeks denoted this as the geometric mean of the lengths AX and AB
(Fig. 2.1):
If we denote the length of the longer ray AB as a, the length of the dis-
tance between X and B as x, and the length of the shorter rays AF, or AX
as b=ax, then a/x=x/(ax). The length x is then the geometric mean of
a and ax, and is also the root to the polynomial obtained from equating
the ratios a/x and x/(ax). Cross-multiplying, we can express these ratios
as x2a(ax). Then:
x is the root of x 2 + ax a 2 = 0.
The ancient Greeks had learned to construct various polygons from the
square, and the series of even-sided polygons that were multiples of the
four-sided figure and its successors. They could also do the same for the
triangle and its even-numbered multiples. Their first challenge, though,
was to form a pentagon.
Such a five-sided polygon had mystic charms, as had its relative, the
five-sided star. The pentagram was the mystic symbol of the Pythagorean
brotherhood. The Greeks showed that the length of the sides of such
regular polygons enclosed in a circle of unit radius could be expressed
with ratios of integers and their square roots (or geometric means). For
instance, a pentagon within a circle can be constructed as five identical
equilateral polygons much like five pieces of an evenly cut pie, with the
round edges squared-off. Such polygons might look as in Fig. 2.2.
We learn in high school trigonometry class that if we divide the 360
of the circle into five identical parts of 72 each, then the width of the
first such triangle enclosed in a unit circle is represented by the distance
(
from the origin to the point A, or cos(360/5)=cos(72)= 5 - 1 / 4. )
The Greeks knew how to construct a unit circle with a compass, and
they could find the length of the square root of 5 by observing that
its value was simply the geometric mean between 5 and 1. In fact, the
Greeks were able to show that they could use only a straightedge and
compass, or, equivalently, with the tools of addition, subtraction, ratios
and square roots, formed from the congruencies in triangles and formu-
which perplexed Ludvig Lorenz (18 January 18299 June 1891), and
Albert Abraham Michelson (19 December 18529 May 1931) and
Edward Williams Morley (29 January 183824 February 1923), among
others, and, by casting the nature of the space-time relationship in a dif-
ferent light, completely recast classical physics, Gauss awoke one morn-
ing while on holiday in Brunswick with a solution in mind to an equally
baffling problem. How he dealt with his first important discovery became
the template for his mathematical pragmatism over a lifetime. And his
often nonchalant confidence and intellectual dismissiveness that followed
also shed light on why some attribute to others the legitimate discoveries
he had made.
the princes of science of their day. In Baghdad in the early ninth century,
the caliph al-Mamun was the patron of a group of learned men known
as the House of Wisdom. Al-Khwarizmi (780850) had developed solu-
tions to quadratic equations but restricted his solutions to those that
yielded positive numbers. The negative and imaginary roots for which we
are accustomed in high school were discarded as nonsensical.
Three centuries later, the Latin translation by Gerard of Cremona
(111487) of al-Khwarizmis Algebra came to the attention of Leonardo
da Pisa (11701250). Leonardo was asked to determine the roots of a
simple cubic equation x3+2x2+10x=20. This is a specific version of the
general form x3+ax2+bx+c=0, which can be shown to be reducible to
a simpler equation:
z 3 + pz + q = 0,
The young Gauss would have knowledge of the utility of the real num-
ber line and of imaginary numbers. Like Wallis more than a century ear-
lier with his negative numbers, Gauss was the first to offer a geometric
intuition to complex numbers that created substance out of the imagi-
nary number line many still regarded as a mathematical oddity, despite
their pleasing properties.
The 18-year-old Gauss apparently did not know that the Norwegian
mathematician and mapmaker Caspar Wessel (8 June 174525 March
1818) had also offered such a geometric interpretation just a few years
earlier, in 1799. Like Wallis, Wessel was also a mapmaker who studied
directions, and hence vectors. It was a natural extension to consider the
real-imaginary coordinate system rather than the conventional real-real
coordinate system we all observe on two-dimensional maps and in the
ubiquitous x-y graphs. But, his paper, Om directionens analytiske betegn-
ing, which he presented to the Royal Danish Academy of Sciences and
Letters in 1797, went largely unnoticed and untranslated until 1897.
Buried in the Danish paper was the concept we use today to add vec-
tors. Wessel stated in his On the Analytical Representation of Direction
that:
2 The Times ofCarl Friedrich Gauss 23
Two straight lines are added if we unite them in such a way that the second
line begins where the first one ends and then pass a straight line from the
first to the last point of the united lines. This line is the sum of the united
lines.
Wessel also applied his notion of vectors and vector addition to the
complex plane. However, it was left to Gauss to bring the concept of the
complex plane to light in the academic world, and to offer a vivid geo-
metric interpretation with powerful application ever since. He did so as
an 18-year-old youth who was trying to solve a problem that perplexed
mathematicians for more than two millennia.
Gauss had not set out to legitimize imaginary numbers, nor to define
the complex plane. He merely discovered a practical intellectual frame-
work that would allow him to solve a multi-millennial dilemma. Indeed,
he would subsequently have discussions with others who recognized the
value of the new analytic geometry he discovered in 1796. But while
Gauss used his results for his own purposes, he did not publish them
until 1831. It was then that he proposed the new term complex number.
He described it thus:
If this subject has hitherto been considered from the wrong viewpoint and
thus enveloped in mystery and surrounded by darkness, it is largely an
unsuitable terminology which should be blamed. Had +1, 1 and -1 ,
instead of being called positive, negative and imaginary (or worse still,
impossible) unity, been given the names say, of direct, inverse and lateral
unity, there would hardly have been any scope for such obscurity.3
Gauss discovery offered the first bridge between algebra and analytic
geometry. By harkening back to the Pythagorean principle that geometric
figures be drawn only with a compass and a straightedge, he also rein-
forced his fascination with the geometric mean, a property that would
prove influential in his development of the least squares methodology.
To see his insight, consider the consequence of drawing vectors and
geometric figures on a complex plane. In such a representation, with the
algebra first proposed by Leonard Euler a half century earlier, and the
representation on the complex plane proposed by Caspar Wessel, the
24 The Econometricians
horizontal axis is the traditional real number line, and the vertical axis
was in units of plus or minus the imaginary number i. In such a complex
number plane, a given point is then described by a real part a and an
imaginary part bi (Fig. 2.4).
The algebra of imaginary numbers had been fully explored by Euler
and others, but the geometric interpretation was novel. The complex
algebra had a number of nice properties. For instance, lets begin with the
simplest statement of exponential and trigonometric equivalency.
Recall de Moivres identity that was used by the great Leonhard Euler
(170783) to subsequently determine, in 1748, that eix=cos(x)+i * sin(x).
The 1965 Nobel Prize-winning physicist Richard Phillips Feynman (11
May 191815 February 1988) labeled Eulers formula the most remark-
able formula in mathematics.4 Euler had been exploring the infinite series
that represent the exponential and then the two trigonometric functions:
xn x2 x3
ex = =1+ x + + + for all x
n=0 n! 2! 3!
He observed the similarity between the infinite sum above and those
of the sine and cosine functions:
( -1)
n
x3 x5
sin x = x 2 n +1 = x - + - for all x
n=0 ( 2n + 1)! 3! 5!
( -1) 2 n
n
x2 x4
cos x = x =1- + - for all x
n = 0 ( 2 n )! 2! 4!
He noted that the infinite series terms for eix would equal the sum of
the terms for cos(x)+isin(x). From this observation, he concluded Eulers
identity:
e ix = cos ( x ) + i sin ( x ) .
ix = ln ( cos x + i sin ( x ) ) .
It is possible that Cotes failed to observe that sin(x) and cos(x) are peri-
odic functions that cycle continuously between the values of 1 and 1 as
x increases. When, in 1740, Euler instead expressed each side as an expo-
nential, he was left with his familiar Eulers formula, which he immedi-
ately recognized as necessarily periodic. Neither discoverers offered the
26 The Econometricians
now familiar interpretation offered by Wessel and Gauss, even though the
interpretation is quite conventional today.
Interestingly, Cotes shared with Gauss a vocation in astronomy and,
like Gauss, was interested in how observation errors tend to regress with
increased observations, rather than multiply. His interest in predicting
the movement of planets, given observational error, was the motivation
for what would eventually result in the method of least squares.
With this history at hand, let us now explore Gauss insights. As
asserted by Gauss, a complex number can be represented as the sum of
a real component and an imaginary component representing the sum of
two vectors on the complex plane. This vector is the sum of the move-
ment in the real direction and imaginary direction. Thus they sum to
a+bi. The parameters a and b then represent distances along the real
horizontal axis and the imaginary vertical axis (Fig. 2.5).
In polar coordinates, the distance along the horizontal axis is just as we
find for the real plane:
x = r cos ( Q )
y = r sin ( Q ) ,
where r is equal to 1in the case of the unit circle. In general, the real
parameters (x, y) of a circle follow the identical equation for a circle as in
the real plane:
r 2 = x 2 + y2 .
Fig. 2.6 Regular polygons in the unit circle on the complex plane
Notice that, for the triangle, the first apex is given by real and imagi-
nary coordinates (1,0), or z0=ei*0, the second apex is z1=ei*2/3 and the
third apex by z2=ei*4/3. Each apex is rotated by 120, or 2/3 radians.
Since complex number multiplication for a unit ray simply represents a
rotation, each apex is simply the square, or the cube, or the kth power of
the first apex. N rotations for an n-sided polygon is then given by zn=1
=en=cos(n)+isin(n). Then, this description of an n-sided regular
polygon developed by Gauss immediately yields De Moivres formula:
( cos ( Q ) + i sin ( Q ) )
n
= cos ( nQ ) + i sin ( nQ ) .
has n roots. If we accept such complex roots, the nth root of unity prob-
lem and its relationship to polygons are immediately apparent.
Gauss next task was to demonstrate that some of these roots can be
described using numbers represented by ratios of whole numbers or their
square roots, or the so-called constructible polygons that can be drawn
only with a compass and an unmarked straightedge. The Greeks had
known they could do so for polygons with 3, 4, 5, 6, 8, 10 and 12 sides.
Each of these are what we call a Fermat prime number, or a multiple of a
Fermat prime number. The next Fermat prime number in the sequence is
17, where a Fermat prime is given by:
n
Fn = 22 + 1
2p
16 cos = -1 + 17 + 34 - 2 17
17
+ 2 17 + 3 17 - 34 - 2 17 - 2 34 + 2 17 .
While it took Gauss a couple of years to write his dissertation into a trea-
tise that proved the assertion he made in 1796, and another three years to
rewrite it in the form of the published treatise Disquisitiones Arithmeticae,
in Latin, in 1801,5 he nonetheless had signaled to the mathematical
world his brilliance in solving a 2000-year-old problem.
In Gauss first year at the University of Gttingen, he was never fully
secure in his personal finances. When he first entered the Collegium in
Brunswick, his funding from the Duke was sufficient, but not permanent.
He was overjoyed when the Duke agreed to fund his first year of study
at Gttingen, but he remained concerned the funding would continue.
Having solved a 2000-year-old problem with an incredibly elegant and
profound solution, the 18-year-old Gauss mathematical credentials were
30 The Econometricians
Perhaps the book that 25-year-old Carl Gauss finally published years
after his Gttingen thesis might have been more quickly written and eas-
ily absorbed had he not written it in Latin. Indeed, his work was one of
the last major works among mathematicians to be written in Latin. The
language was one still studied by academics of all nations and hence his
papers could be read among a wide subset of a very narrow circle of elite
mathematicians. Soon thereafter, English would be adopted as the uni-
versal language of academicians.
Indeed, his treatise might have been even longer and more expansive,
had finances not stood in his way. It was the custom then that the inves-
tigator, or his patron, to pay to have their books published. To reduce
these costs, Gauss measured his contributions not by the expansiveness
of his ideas but by the economy of his brevity. His expositions were terse,
to the point that many lesser minds could not fill in the gaps Gauss left.
And, he had trimmed a chapter from the book to shorten it and reduce
its publication costs. When the Duke of Brunswick eventually discovered
Gauss fiscal plight, he offered to fund the completion of the book, which
was finally published in 1801, two years following the completion of his
dissertation.
The most brilliant mathematicians of Gauss day quickly understood
the contribution, though. Lagrange noted that the book immediately
placed Gauss within the highest echelon of mathematicians.
Notes
1. H.all, Tord, translated by Albert Froderberg, Carl Friedrich Gauss, MIT
Press, Cambridge, Massachusetts, 1970, p.21.
2. Merino, Orlando, A Short History of Complex Numbers, January, 2006,
http://www.math.uri.edu/~merino/spring06/mth562/ShortHistory
ComplexNumbers2006.pdf
3. Nahin, Paul J., An Imaginary Tale, Princeton University Press, Princeton, NJ,
1999, at p.61.
4. Feynman, Richard P. (1977). The Feynman Lectures on Physics, vol.
I.Addison-Wesley. pp.2210.
5. Carl Friedrich Gauss, Carl Friedrich, translated by Arthur A.Clarke,
Disquisitiones Arithmeticae, Yale University Press, 1965.
3
Carl Gauss Great Idea
Johanna was the star of her parents eyes. She was an only child,
cheerful and kind, but rather unsophisticated. She offered Gauss some-
one he described in a letter as a perfect life companion. Gauss courted
Johanna for a year before he professed his undying love to her and asked
her to marry him. It took her four months and ten days to finally agree
to his request. She feared his fame and his worldly life would eclipse her
humble desires. Yet, less than a year later, on 9 October 1805, they mar-
ried at the same St. Katharines Church that adjoined the school annex
where he first impressed his teachers with his mathematical skills.
The young couple had a child less than a year later, on 21 August
1806. They christened their son Joseph on 24 August, at St. Katharines
Church.
Things were looking up for Gauss in 1806. Based on his astronomical
celebrity, the Duke had bought Gauss a state-of-the-art telescope and
Gauss had been instilled as Brunswicks resident astronomer and observa-
tory director. But, despite these joys, tragedy almost immediately beset
Brunswick and his patron. Gauss hometown became embroiled in the
Napoleonic Wars, and their Duke was enlisted as a combat general allied
with Prussia. An attack of his army by Napoleons forces caused the Duke
to receive a blinding wound on the battlefield. He was permitted to
retreat to Brunswick, but was disgraced as Napoleon overtook his town.
The Duke was driven into exile, and soon died, on 10 November 1806.
Meanwhile, Gauss had been working as the director of Brunswicks
observatory. When Napoleons regime tried to extract a war contribution
of 2000 francs from him, numerous scientific luminaries each sent the
sum to Gauss, which he promptly returned. With the loss of his beloved
patron, his financial pride bruised and his fear of Napoleons court well
founded, Gauss was miserable in Brunswick. Johanna was unhappy, too,
despite the arrival of a daughter, Wilhelmine, on 29 February 1808. Six
weeks later, Gauss father died in Brunswick. Gauss patron removed,
and with some jealousy among his fellow residents for his privileged sta-
tus within Napoleons court, Gauss eventually left Brunswick. The Gauss
family moved to Gttingen on 21 November 1807.
Yet, despite these sorrowful distractions and melancholy, Gauss
remained productive. In 1809, he published his Theoria Motus Corporum
Coelestium in Sectionibus Conicis Solem Ambientium,1 which presented his
3 Carl Gauss Great Idea 37
method for describing the motion of orbiting bodies and his method of
least squares. With its publication, Gauss became a member of learned
societies around the world. In 1809 and 1810, his renown accelerated,
and he received numerous offers for professorships.
Meanwhile, he and Johanna had their third child, Ludwig, on 10
September 1809. He died less than six months later, on 1 March 1810.
Preceding him in death was Johanna, on 11 October 1809, as a conse-
quence of two difficult childbirths in a little more than a year.
Gauss was grief-stricken over the loss of his father, his patron, the
Duke, his wife and his son in less than four years, and remained bitter of
France over its treatment of Germany. Nonetheless, he vowed to remain
in Gttingen, to staff its observatory and oversee the construction of a
new state-of-the-art observatory. From his vantage point in Gttingen,
Gauss observed the Great Comet of 1811 and again successfully calcu-
lated its orbit using his method of least squares. One of the brightest
comets ever observed, it captivated the attention of citizens worldwide.
Some believed it portended to Napoleons invasion of Russia, and the
War of 1812 between the USA and Great Britain.
While the constructability of the 17-gon harkened Gauss arrival
among the mathematical elites, his method of least squares cemented
his position in the minds of practical scientists and astronomers. The
prediction of the movement of celestial bodies, and the study of geod-
esy, or the understanding of the applied mathematics of the shape and
orientation of the earth, was essential for the Age of Exploration within
which Gauss lived. Navigation between continents without the aid of
land sightings required mechanisms to correct the random measurement
errors of navigators.
Men of science, from Galileo to Cotes to Laplace, had offered ways
to tease accuracy out of observations subject to random error, by the
averaging of observations. Laplace had perhaps the best success when he
augmented the work of the physicist and mathematician Roger Joseph
Boscovich (18 May 171113 February 1787) in the method of least
absolute deviation to discern the true observation when observational
errors exist.
The term observational errors that we still use today suggests the root of the
original problem. These errors were not a mere scientific annoyance. Because
38 The Econometricians
It is the last point that created the greatest discussion. If one simply
aggregated the data to calculate its mean, positive errors would cancel out
negative errors, and large deviations would overwhelm small ones. Galileo
proposed the use of the sum of absolute errors to overcome some of this
concern. As such, positive deviations of observations from a hypothesis
of the true value would not cancel out negative ones. Then, the researcher
could propose a hypothesis that would minimize the sum of the absolute
value of errors.
Galileos approach is consistent with the concept of Occams Razor.
Named for William of Ockham (12871347), a Franciscan friar and
philosopher living in England. It is often stated as Among competing
hypotheses, the one with the fewest assumptions should be selected. All
else remaining the same, the measurement that best fits the data should
be accepted.
Yet, despite this logic, and perhaps because Galileo did not offer a full
solution to the dilemma, the simplest approach was to simply average the
observations. Roger Cotes, Sir Isaac Newtons contemporary and aide,
40 The Econometricians
abs( x )
1
p( x) = e b
2b
The issue may nonetheless be settled. Gauss had already been completing
a major treatise on the subject when Legendre published his own. While
the translation from Latin, and the need for less affluent Gauss to raise
the funds for publication, delayed its appearance until 1809, when he
managed to publish his Theory of the Motion of Heavenly Bodies Moving
about the Sun in Conic Sections,4 Gauss clearly demonstrated a much more
extensive command of the method of least squares than Legendre had
described. What we know of the method, including the result that it mini-
mizes the sum of squared errors when errors follow a normal distribution,
flows from this publication by Gauss.
Perhaps part of Gauss indignation arose because of the immediate rec-
ognition Legendre received upon his publication. Meanwhile, Gauss had
never published his own method that so accurately predicted the reappear-
ance of Ceres, but obviously could not have made his prediction without
the application of such a technique. Certainly, his principle of probability
and of the normal distribution provided a much more extensive founda-
tion than the method Legendre derived, and cemented Gauss reputation.
In his treatment, Gauss began with the work of Laplace decades ear-
lier. He was the first to demonstrate that the probability density function
he derived was an evolution of Laplaces solution. Rather than the mini-
mization of absolute deviation, Gauss treatment minimized the sum of
squared errors in estimation. From the premise that the arithmetic mean
of observations should provide an unbiased estimate of the true mean
parameter, Gauss then demonstrated that, if observation errors are nor-
mally distributed, his method of least squares provides the most likely
estimate of the true measurements. Our understanding of the nature of
errors was expanded in the two years following Gauss publication of his
method of least squares, to be described later. From these principles Gauss
formulated, Laplace completed the analyses we use today by proving the
central limit theorem, something that had stymied Laplace since 1783.
In 1810, following a close study of Gauss work, Laplace showed that
the central limit theorem offered a Bayesian (posterior) justification for
Gauss least squares methodology. If observations are combined, each of
which is an unbiased and independent observation drawn from a large
number of observations, then the least squares estimation represents the
maximum likelihood estimate and minimizes posterior errors, all with-
46 The Econometricians
out any assumption with regard to the true distribution of errors. This
central limit theorem proved that the sum of a number of independent
random variables that are identically distributed will tend to a normal
distribution
Gauss further extended his model by showing that the least squares
approach to a linear regression is optimal if errors have zero mean and
are uncorrelated, with equal variances. This result is known as the Gauss-
Markov theorem. He also derived the now ubiquitous normal distribu-
tion from first principles.
Meanwhile, in the USA, Robert Adrain (30 September 177510
August 1843), considered to be the pre-eminent American mathemati-
cian of his day, formulated a similar but less broad analysis as had Gauss.
n
P ( k successes in n rounds ) = p k (1 p )
nk
k
n
where the terms for n and k in brackets, k , is a binomial coefficient.
This coefficient, read aloud as n choose k, was calculated by Pascal, for
which he derived what is now called Pascals triangle to aid in its calcula-
tion. Jacob Bernoulli was able to approximate the binomial coefficient
48 The Econometricians
well beyond the estimates provided by Pascal and Fermat, but was unable
to determine an approximation that was easy to compute. It was left to
De Moivre to calculate, in 1733, the probability of coming within d out-
comes half the time of n repetitions given even (p=0.5) odds as:
n
nn 1 4 2
d = e 2 d / n
2 2 2 n
n
min f ( x ) = min x xi
x x
i =1
It is well known that this minimization yields the median rather than
the average value of the set of observations xi. We had also noted earlier
3 Carl Gauss Great Idea 49
that Cotes had argued for a center of mass calculation such that each
observation is weighted by its distance from the calculated center. Under
such a formulation, if each observation is given a weight wi, and these
weights sum up to one, then the center of observations x occurs such that:
w ( x x ) = 0.
i i
i =1
w x i i
x= i =1
n
.
xi
i =1
This expression reduces to a simple average when the weights are equal.
The issue of the appropriate adjustment for repeated observations that
contain a random component was most pressing. There was even testi-
mony to Parliament on the importance of resolving this issue.
Laplace turned his attention to the appropriate error distribution and
proposed a symmetric decay function for the probability weighting of
observations from the central tendency. He argued that the effect of a small
change in one direction or the other, when compared to the central ten-
dency, should equal the proportion of the change to that tendency. From an
argument of constant proportional change as one departs from the central
tendency, he derived the exponential decay function, which, in his formula-
tion, depended on the absolute distance of an observation from the central
tendency. The weighting, or probability assigned to a deviation, then decays
exponentially and symmetrically on either side of the central tendency.
Laplaces more elaborate work on the error function convinced few,
though. Daniel Bernoulli (8 February 170017 March 1782 ), the
nephew of Jacob Bernoulli, lent his considerable support in 1777 to
Cotes notion of the center of mass, or simple weighted averaging, under
the assumption that all errors are equally likely, despite the pleadings a
century earlier of Galileo who argued that small discrepancies are likely
more probable than large ones, and hence should hold more weight.
50 The Econometricians
Bernoulli was also offended by the explicit notion of Laplace that errors
can occur over an infinite line. To Bernoulli, it was common sense that
possible errors should be finite or asymptotically declining in nature.
Yet, averaging, or choosing the median among observations, did not allow
astronomers to calculate the reappearance of Ceres in 1801. Gauss, in advo-
cating for a least squares methodology and his eventual normal distribution,
noted that, as Galileo claimed, small errors should be more likely than large
ones, and the errors should occur symmetrically about the true central ten-
dency. In addition, the most likely occurrences, in a probabilistic sense, ought
to coincide with their average. From this, he concluded that departures from
the mean ought to follow an error curve that does not look significantly dif-
ferent from the approximation to the Bernoulli trial De Moivre had postu-
lated. Gauss asserted an error function (x) should be given by:
h
( x) =
2 2
eh x ,
where h is a precision constant that ensures the sum of these probabilities
along the real number line sum to 1. Note that, if the true mean is given
by 0, and the standard deviation is given by 1 / 2h , the expression
reduces to our familiar one for the normal distribution.
To see how Gauss arrived at this conclusion, let us revisit his goal. He
wished to discover a probability distribution function that reaches a maxi-
mum when the observation x equals the true mean . The joint probabil-
ity density for n observations of x is then given by the following product:
n
x : = ( xi )
1
Gauss wanted his distribution to peak when the mean of the set of obser-
vations coincides with the true mean . He had developed a maximum
likelihood methodology to determine the mean. Differentiating the product
with respect to the true mean and setting the derivative to zero then gives:
n
( xi : )
0 = (x : )
i =1 ( xi : )
3 Carl Gauss Great Idea 51
Let zi=xi, and ( x : ) = ( x : ) / ( xi : ) . Then, if the mean of
i
n n
0 = zi and 0 = ( zi ) .
i =1 i =1
d / dz
= kz and d = kzdz.
where k=1/2, and where we note that the exponent in the integrand
must be negative to ensure the integral remains finite over the entire
range of values for x.
Finally, we can determine the arbitrary constant C by noting that the
sum of all probabilities must be equal to one:
2
1 ( x )
2
1 = Ce dx.
1
( ( x ) / )
2
h 1
C= , ( x) = e 2
.
2
b 1 x / 2
1 ( ( ) )
Prob ( a < x < b ) = e 2
dx.
2 a
1
( z / )
2
1
Var ( z ) = z e
2 2
dx.
2
2 2
Var ( z ) = we
w2
dx.
2 2 w w2 1 w2
Var ( z ) =
2 e 2 e dw .
+
2 2 1 w2
Var ( z ) = e dw .
2
2 2
Var ( z ) = = 2 .
2
3 Carl Gauss Great Idea 53
The variance of the error terms is then given by 2. The familiar normal
distribution was derived based on Gauss premises that the distribution
maximum, the average and the median of the probability density func-
tion he derived all occur at the true mean . He also showed that his dis-
tribution has a variance of 2 and illustrated it has the familiar bell shape
we now know. Purists still call it the Gaussian distribution (Fig. 3.1).
The next innovation came at the hands of Laplace. Over the years,
Laplace had closely followed Gauss maturation as a mathematician, and
had remained one of his greatest supporters. Laplace had even intervened
to have Frances science academy award Gauss their highest honor, and
ensured that Napoleons Army treated Gauss well as it rampaged through
Europe, Brunswick included. When the nasty row over priority erupted
between Gauss and Legendre, Laplace tried to smooth it over. And, when
Gauss definitive treatment of the method of least squares was finally
printed in 1809, Laplace heralded its contribution and set about study-
ing its results. From there flowed one of the most important innovations
in statistic: the central limit theorem. Laplace first described it in his
1812 Thorie analytique des probabilits, very shortly after Gauss publica-
tion of his unifying and extensive work in 1809.
To understand the significance of the central limit theorem, it is help-
ful to note a characteristic of the mean and variance we just derived. As
you recall, the mean is simply the expected value of a random variable,
= E(X )
= Var ( X ) = E ( X 2 ) 2 .
2
These are called moments because, like the moments of inertia relative
to a center of mass, they are measures of the distance of various points
from the central tendency, weighted by their probability, or frequency, or,
in the case of physical objects, their mass. In statistics, it is often easier
to work with a moment-generating function to determine these various
parameters as moments. Then, the mean of a random variable, or the first
moment, is simply the first derivative of the moment-generating func-
tion, and the variance the second derivative, less the mean-squared, or
2, and so on. Specifically, the moment-generating function is given by:
( )
M ( t ) = E etX = etx f ( x ) .
xS
n
t
( )
n
M Sn ( t ) = M xn ( t ) and M Z n ( t ) = M xn .
n
x
1
M x ( s ) = M x ( 0 ) + sM x ( 0 ) + s 2 M x ( 0 ) + s ,
2
1
M x ( s ) = 1 + s 2 x2 + s .
2
n
n t
2
1 t 2
+ n s
M Zn ( t ) = 1 + x2 + s = 1 + 2 .
2 x n n
n
a
lim 1 + n = e a .
n
n
Then we have:
n
t
2
+ n s
lim M Z n ( t ) = lim 1 + 2
2
= et / 2 .
n n
n
56 The Econometricians
Notes
1. GAUSS, Carl Friedrich (17771855). Theoria motus corporum coeles-
tium in sectionibus conicis solem ambientium. Hamburg: Friedrich
Perthes and I.H.Besser, 1809.
2. Galilei, Galileo (1632), Dialogue Concerning the Two Chief World
System, translated by Stillman Drake. Berkeley, CA: University of
California Press, 1953.
3. Legendre, Adrien-Marie (1805), Nouvelles mthodes pour la dtermi-
nation des orbites des comtes [New Methods for the Determination of
the Orbits of Comets] (in French), Paris: F.Didot.
4. Gauss, Carl Friedrich, (1809), Theoria Motus Corporum Coelestium
in sectionibus conicis solem ambientium (Theorie der Bewegung der
Himmelskrper, die Sonne in Kegelschnitten umkreisen), Theory of
the Motion of Heavenly Bodies Moving about the Sun in Conic Sections
(English translation by C.H.Davis), reprinted 1963, Dover, NewYork.
4
The Later Years andLegacy ofCarl
Friedrich Gauss
If a line segment intersects two straight lines forming two interior angles
on the same side that sum to less than two right angles, then the two lines,
if extended indefinitely, meet on that side on which the angles sum to less
than two right angles.
It is not knowledge, but the act of learning, not possession but the act of
getting there, which grants the greatest enjoyment. When I have clarified
and exhausted a subject, then I turn away from it, in order to go into dark-
ness again I imagine the world conqueror must feel thus, who, after one
kingdom is scarcely conquered, stretches out his arms for others.
Note
1. https://math.dartmouth.edu/archive/m5w00/public_html/quotes.
html, accessed 18 January 2016.
Part 2
From Least Squares to Eugenics
Elizabeth COLLIER
(COLYEAR)
b: 1713
d:
178330 March 1844) took over the family business in 1815, the family
had become firmly established as bankers.
Samuel Galton Jr. died in 1832 a wealthy man, with a large fortune
estimated at 300,000. By then, his son Samuel Tertius, a well-educated
graduate of Cambridge University, had increasingly lived a life of nobility.
70 The Econometricians
Perhaps imbued with his half cousins love for exploration, Galton
took to adventures of his own. He sought new experiences, first in Egypt,
and then throughout the Middle East, accompanied by a servant and by
friends he met there, or whom would join him from England. He eventu-
ally bored of his adventures, though, and returned to the life of country
gentry in England.
Soon, though, he again yearned to travel, and set out to travel even
more extensively. Much of the latter half of the 1840s would be his age
of exploration of Africa. Upon his return in the early 1850s, he was
celebrated for his adventures. His writings and subsequent speeches to
learned societies on his African travels won him respect in Londons intel-
lectual community and the major award from the Royal Geographical
Society that David Livingston, of Dr. Livingston, I presume notoriety,
had just won some time earlier for his own explorations of Africa.
His reputation as a noted adventurer cemented, Galton traveled his
circles in London and beyond, often as the toast of the party. On one
occasion, in the Christmas holiday season of 1852, he attended a party
at his neighbors home and met his future wife, Louisa Butler, the daugh-
ter of George Butler (5 July 177430 April 1853), a well-known math-
ematician from Cambridge, and sister of other Cambridge students and
alumni. Galton fit well into the family. Francis and Louisa were married
on the 1 August 1853.
A marriage-blissful Francis set about to work on a book of his adven-
tures in Africa. His publication became a popular read, and more awards
were bestowed upon him. Barely 32 years old, Francis had achieved as
much fame as a man of leisure could earn. He augmented his fame with
additional highly successful books, and with a lecture series. By 1856,
he was appointed a Fellow to the Royal Geographical Society. From his
learned base, he gained the friendship of Englands most influential think-
ers of the day. Among the dozens of intellectuals he came to know was
Herbert Simon, the father of Social Darwinism, and the individual who
coined the phrase survival of the fittest to so compactly, and perhaps erro-
neously, as a description of the contribution of Galtons cousin, Charles
Darwin. This troika of Galton, Darwin and Simon collectively began
to revolutionize humankinds thought on evolutionary destiny from the
perspective of both biology and society.
6
The Times ofFrancis Galton
were larger than average did not produce offspring that were larger yet.
Nor would smaller parent seeds produce offspring that were smaller yet.
Instead, he established that the weight of offspring seed vary from the
mean, on average by only one-third of the deviation from the mean of
the weight of the parent seed.
In interpreting this peculiar relationship, Galton said that offspring
regress to describe this tendency toward the mean, and coined such an
analytic description a regression. The one-third rate was described as the
regression coefficient. Some sort of natural process then appeared to
Galton to dampen extremes and causes offspring to regress toward the
mean.
Galton argued that his technique could actually be used to compare
many such interactions. Of course, there is no problem with units when
on both axes is a measure of height. The slope of a line that compares one
axis to another is a ratio without units, or, if one prefers, a rate of inches
to inches. In other circumstances, one might posit that rainfall might
influence a crop yield, or, in the finance literature, increased risk might
command a greater return, as in the capital asset pricing model (CAPM).
In these cases, the correlation between one variable and another might
better be described by some sort of correlation coefficient rather than a
unitless slope.
It was this statistical extension that Galton set out to establish using his
regression model adapted from Gauss ordinary least squares methodol-
ogy. To create the data for his analyses, he established the Anthropometric
Laboratory in nearby South Kensington to solicit subjects who would be
willing to be measured in a multitude of ways. In his lab, Galton and his
researchers established the first extensive database of human measure-
ments and qualities, and even pioneered such lifelong markers as the use
of fingerprints.
To some, Galtons audaciousness and academic entrepreneurship were
little more than self-aggrandizement, at best, and derivative at worse.
Others heralded him as a genius. There was one certainty, though. Galton
always walked along the cliff of controversy. Despite his confidence that
he was an exceptional polymath, he was in fact not particularly skilled
mathematically.
80 The Econometricians
y = r ( sx / sy ) ,
where r is the slope of the graph when the two variables are plotted
against each other.
While Galtons analysis was incomplete and non-rigorous, he none-
theless provided a lasting intuition, beyond his coining of the expression
regression. He correctly observed that the relative variability of two fac-
tors was an important determinant to the scale of their relative graphical
depictions, and hence the slope of the regression line. He would leave to
one of his laboratory assistants, his prodigy Karl Pearson, to formalize,
and hence legitimize his intuition.
Note
1. Galton, F., Hereditary Genius, Macmillan, London, 1869.
7
The Later Life andLegacy ofSir Francis
Galton
One might contrast the life of Gauss with that of Galton. Gauss humble
beginnings might suggest he had everything to prove. Yet, over his life-
time, and despite his place as perhaps one of the three most accomplished
mathematicians of all time, he took far too little time documenting and
publishing his contributions. His brilliance was understated in his own
lifetime.
On the contrary, Sir Francis Galton was born to the purest of pedigrees
and privilege. The cousin of Charles Darwin and a member of a family
of famous physicians, well-healed bankers and prominent theologians,
Galton was a bigger than life personality from an equally impressive fam-
ily. He had little hed need to earn, but spent a lifetime trying to establish
and enhance his reputation. His brilliance was equally overstated, as per-
haps were the accolades he received. Yet, he left a prodigy, Karl Pearson,
for whom he endowed an academic chair and hence a livelihood. Pearson
spent his career filling in the academic blanks Galton had left.
While Galton cultivated a perception that he was a polymath, his
prodigy certainly was.
Thomas Pearson
b: 1797 in Crambe, Yorkshire,
England
m: 1818 in Yorkshire, England Elizabeth Hopkinson
d: May 1859 in Yorkshire, England b: Abt. 1748
d: 05 Jun 1831
William Pearson
b: 28 Oct 1822 in Crambe,
Yorkshire, England
m: 29 Dec 1853 in Stepney,
Middlesex, England Richard Blacksmith Beilby
d: 15 Oct 1907 in Hampstead, b: 18 Feb 1766 in Crambe,
Middlesex, England Yorkshire, England
m: 17 Aug 1790 in North Grimston,
Yorkshire, England
Elizabeth Beilby d: 16 Aug 1816 in North Grimston,
b: 1800 in Yorkshire, England Yorkshire, England
d: May 1882
Ann Wray
b: 1776
d:
Carl Pearson
b: 27 Mar 1857 in Islington,
Middlesex, England
m: 30 Jun 1890 Smith
d: 27 Apr 1936 in Capel, England b:
m:
d:
Thomas Smith
b: 1782
Fanny Smith m:
b: 30 May 1827 in Hull d: 1863
d: 1905
By then, his brother had received a scholarship in the study of the clas-
sics at Trinity College, Cambridge. Carls father wanted at least one of
his children to study mathematics, so he secured a Cambridge Wrangler
to prepare Carl for the Tripos entrance exam. Wranglers are those who
receive the highest honors in their third year studies at Cambridge. This
honor of Mathematics Wrangler connotes a graduate of intellectual
supremacy. Carl was sent to the town of Hitchin, near Cambridge, for
five months of intensive mathematical tutoring by a Wrangler in early
1874. Unhappy in Hitchin, he left that summer to be tutored in math-
ematics at Merton Hall, Cambridge, by a cadre of tutors which included
legendary Wrangler John Routh.
86 The Econometricians
Edward John Routh FRS (20 January 18317 June 1907) was con-
sidered the best among Senior Wranglers in preparing students for the
exam. Born in Quebec, Canada, to a well-established family whose
fathers ancestors also originated in Yorkshire, Routh studied mathemat-
ics at the University of London before he continued on at the Peterhouse
in Cambridge. He, too, was prepared by the senior wrangler maker
William Hopkins FRS (2 February 179313 October 1866). Routh
himself graduated as a Senior (signifying top) Wrangler, just ahead of
the brilliant Scottish mathematical physicist James Clerk Maxwell FRS
FRSE (13 June 18315 November 1879). Clearly, young Carl could not
be in better academic hands.
Routh instructed Carl in a mathematics that was heavily laden with
physics. Over the next nine months, Carl studied in preparation of the
exam, which permitted him a scholarship at Kings College in Cambridge.
He began his University studies on 9 October 1875.
Carl Pearson thrived in this intensive intellectual environment. As a
child he was somewhat frail and sickly, and he felt a lack of warmth, car-
ing and inclusiveness. Kings College believed in exercising the mind and
the body. Carls emotional and physical constitutions were strengthened,
and he thrived intellectually. He was also immersed in the Classics and
of the Romantic school. He graduated as Third Wrangler in 1879, which
translates to third in his class.
His academic success also earned him a Kings College Fellowship.
This lucrative scholarship allows the recipient up to seven years of fund-
ing, with no teaching expectations, to pursue his research agenda. The
scholarship recognized Carl Pearson as among the most distinguished
and promising university graduates in the nation.
Carl had yet to travel, though. He was not brought up in a wealthy
family as had Francis Galton, and Continental travel was not a luxury
his family could afford. His fellowship made more options available to
him. He had been studying mechanics and engineering, with the hope
of becoming an engineering physicist. Germany was the center of the
study of physics at the time, so he began to study German and traveled
to Heidelberg. While there, he was enticed by the theories of the great
philosophers, from Kant and Spinoza to John Locke.
8 The Early Life ofKarl Pearson 87
Almost immediately after he began, Karl gave up the practice of law. Yet,
the pressures for him to succeed were almost unrelenting. To free him
from the forces of familial conformation, he joined the intellectual circles
of London. He lectured locally at the intellectual clubs in Soho, includ-
ing the Men and Womens Club. While he considered himself a man of
numbers, others increasingly viewed him as a man of words.
From his engineering training, Karl was fascinated by the theory of
elasticity. This is a mathematically rigorous application of the principles
of physics and engineering that governs the bending of materials such as
bridge spans and beams and the forces as objects move through a viscous
medium. Karl also pursued an eclectic combination of studies that was
not only heavily influenced by the mathematics of the day. He was also
fascinated with the philosophy of science that imposed on its practitio-
ners the need to look at familiar problems in unfamiliar ways.
Karl Pearson was especially influenced by a mathematical physicist
and philosopher William Kingdon Clifford FRS (4 May 18453 March
1879), a brilliant non-Euclidean geometer who argued for the equiva-
lence of mass and energy and the notion of the curvature of space. The
intellectual explorations he began in his 1876 On the Space-Theory of
n
xi yi
n
,
x =i
where x and y are the deviations between the predicted means and the
pairs of data. For instance, in the vertical direction, this would be:
( y y ) .
2
( x )( )
n
i x Yi Y
b= i =1
( x )
n 2
i x
i =1
96 The Econometricians
a = Y bx,
( )
n
( x i x ) Yi Y
r= i =1
.
(Y Y )
n n 2
( x x)
2
i i
i =1 i =1
1 n
y= yi .
n i =1
SStot = ( yi y ) .
2
i
9 Karl Pearsons Great Idea 97
SSreg = ( fi y ) .
2
SSres = ( yi fi ) .
2
Pearsons goodness of fit measure, often now called R2, is then given by:
SSres
R2 1 .
SStot
This framework for goodness of fit can also be employed for the deter-
mination of quality of other predictive models. The variation of this
goodness of fit measure is then called the coefficient of determination.
By Pearsons era, much of the history of the statistics to which a finance
student will be exposed in their first yearlong course in statistics was at
least partially established. Not yet fully explored was the least squares
linear regression model and its properties. Later we describe the linear
regression model and document the important extensions to finance that
were offered by Fisher, Hotelling, Frisch and others.
Note
1. Pearson, Karl, The Grammar of Science, Adam and Charles Black Publishers,
London, 1892.
10
The Later Life andLegacy ofKarl
Pearson
some that should have remained shut, and would have been bolted shut
for men of lesser wealth like Gauss, or Pearson.
Galton paved a path for Pearson, and was perhaps even his academic
benefactor. Each recognized and rode on the others coattails. Certainly,
of the two, Pearson was the more academically brilliant. He also was
more driven to succeed as he realized that, unlike Galton, he would have
to earn each bit of his success.
Pearson could not afford the life of an adventurer as could Galton.
His brief forays in travel resulted in his awakening to the growing social-
ist movement on the continent, the writings of such individuals as John
Locke and Karl Marx, and the brilliant work of Europes leading math-
ematical physicists, mostly in Heidelberg and Gttingen at that time. He
could bring back to London the lofty ideas of these philosophers, but he
felt out of the league of their physicists. His admiration for them seemed
to cause him to forever change the spelling of his first name from Carl,
to the German version Karl, though, like another famous London intel-
lectual at the time, Karl Marx.
In 1885, Karl Pearson found an ideal outlet for his socialist thought,
his procreation ideals along the lines of eugenics, and his desire for poten-
tially feisty companionship. He founded The Men and Womens Club,
with the goal of attracting an equal number of men and women from the
middle and upper middle class who espoused progressive views on social-
ism, feminism and sexuality.
Then a 28-year-old bachelor, Pearson firmly believed that more
empowered and educated women were necessary for national advance-
ment. He gave a paper at the first of the groups monthly meetings near
Soho entitled The Womans Question, in which he espoused greater access
to education, politics and the professions for women, at the time when
the womans suffrage was increasingly discussed. Just 20 years earlier, the
great economist John Stuart Mill (20 May 18068 May 1873) had been
elected to Parliament partly based on his equally progressive views toward
womens right to vote.
Those who believed in womens suffrage were distinct from the suf-
fragettes who would use violent means to achieve the same desired goals.
The men attracted to Pearsons club were drawn from the prevailing liber-
ally oriented mens clubs for London professionals at the time. The vast
10 The Later Life andLegacy ofKarl Pearson 101
majority of the women attracted were single and were teachers or writers.
Only one of the women had attended university. Indeed, university was
quite inaccessible to women of middle-class families in that era.
This group of a couple of dozen men and women met in homes around
the Kensington area but, over the four-year life of the club, ultimately
failed to come to a meeting of the minds between the genders. While
the men favored a post-patriarchal society which empowered women
and which viewed sex in terms beyond procreation, or in the words of
Pearson, a physical pleasure like climbing a mountain, the men of
the group became increasingly fearful of the feminine energy they were
releasing, and as the discussions became stalemated, almost strictly along
gender lines. The club disbanded in 1889, four years after it began.
At the last meeting of the Men and Womens Club, in March of 1889,
Pearson placed his relationship with Francis Galtons ideas in perspec-
tive. He lamented about the dangers of using tools of exact science in
the realm of eugenics or economics. His prescience was profound. An
overemphasis of the powerful tools of the sciences would detract from
the humanity and complexity of real life. To try to attain mathematical
perfection within theories of human inexactness suggested to him math-
ematical zealotry for mathematics sake.
He concluded this personal philosophical exploration with a greater
sense of the need to use the tools of mathematical statistics to at least
inform the better construction of descriptive statistics. Indeed, Pearson
harbored hope that the discipline of science could instead act to improve
the social dialog and interactions to the point that individuals behaved
in a more rational way. If humans could be educated on the scientific
method, perhaps then their interactions could be more appropriately
explored by the tools of modern statistics. In the early part of the decade
of the 1890s, Pearson devoted himself to higher education reform so that
he may help guide the future of Londons great universities. This was a
period of upheaval among these institutions, and Pearson felt he could
play an important role within such reforms. But, his stakes were higher
than simple university reform. He was simultaneously devoted to social
reform.
It was this concept and ideal that motivated Pearsons Grammar of
Science series. And while one series does not make a revolution, Pearson
102 The Econometricians
William Sharpe
b: 27 Jun 1804 in Marylebone,
Middlesex, England
m: 1841
d: 20 May 1870 in Islington,
Middlesex, England
Maria Sharpe
b: 1851 in Islington, Middlesex,
England
m: 30 Jun 1890
Reid
d: 1928 in Hampstead, Middlesex,
England b:
m:
d:
Thomas Whitehead Reid
b: 08 Oct 1786 in Bristol,
Gloucestershire, England
m: 08 Sep 1813 in Exeter, Devon,
England Whitehead
d: 06 Mar 1845 in Hampstead, b:
Middlesex, England d:
Lucy Reid
b: 06 Jul 1814 in Whitechapel,
Middlesex, England
d: 1896 in Islington, Middlesex,
Timothy Kenrick
England
b: 26 Jan 1759 in Ruabon,
Denbighshire, Wales
m: 24 Feb 1786 in Exeter, Devon,
Mary Kenrick England
b: 05 May 1791 in Exeter, Devon, d: 22 Aug 1804 in Exeter, Devon,
England England
d: 1878 in Hampstead, Middlesex,
England Mary Waymouth
b:
d: 04 Nov 1792 in Exeter, Devon,
England
Marias concern that she would lose her independence in their relation-
ship was well founded. She soon found herself abandoning her feminist
researches and immersing herself in raising ideally eugenic children, host-
ing the parties one would expect of someone of Karl Pearsons stature
in London, and otherwise retreating to her own thoughts and marital
responsibilities.
Together, the Pearsons had three children, Sigrid Loetitia in 1892,
Egon in 1895 and Helga Sharpe in 1898. As we shall see, Egon went on
to become a renowned statistician in himself, if perhaps not in his own
right.
104 The Econometricians
Karl Pearson remained the Galton Professor of Eugenics and the head
of the statistics program until his retirement in 1933. He died three years
later, on 27 April 1936. His wife had passed away before he did, on 30
June 1928.
Over his life, Pearson published dozens of papers, including a series of
18 papers in the Philosophical Transactions of the Royal Society, with the
title Mathematical Contributions to the Theory of Evolution, but with
different subtitles relating to evolution, eugenics, sociology, genetics and
anthropology. In these papers, he described his moments approach, the
chi-squared, correlation ratios, multiple regression, scedasticity, coeffi-
cient of variation and standard deviation. He also established the general
use of lowercase Greek letters to describe population parameters. Pearson
claimed to label the Gaussian distribution the normal curve, although
Gauss had already used the expression normal to denote the quality of his
method of least squares that used minimum distances of errors from the
predicted function. Such minimum distances, from a geometrical per-
spective, are denoted by a normal vector, or a line drawn orthogonally
at a 90 angle from the data point to the representative curve. Gauss
geometric perspective also explains the use of squared deviations because
the square root of a sum of squared deviations, in each direction, gives the
total distance of a point to the function.
Overall, Pearson published over 300 papers on theoretical and applied
statistics, social issues such as mental illness, scientific issues drawn from
astronomy, meteorology, civil engineering, and biology, anthropology
and sociology, and philosophical issues. He also worked on a four-volume
book on the life of Francis Galton.
Pearsons papers analyzed the correlation coefficient, his method of
moments, Pearsons system of continuous curves that was the precursor
to the concept of continuous probability distributions, the chi distance
and the P-value, the chi-squared test, a method of curve fitting by mini-
mizing chi distance (called principle component analysis), the coefficient
of racial likeness as a way to classify races based on the shape of their
skulls, and the establishment of foundations of hypothesis testing that
used p-values and was the precursor to type-I and type-II error analysis.
Pearson also co-founded and edited the journal Biometrika, and edited
it until he died, at which time his son Egon took over the editorial
10 The Later Life andLegacy ofKarl Pearson 105
of them as apart if the events are distinguishable from each other. Space
and time thus offer an opportunity to distinguish for an observer events
that coexist. One aspect of the distinctness lies in space, and the other in
the sequence, or time.
Pearson was arguing that time is simply the dimension that allows us
to distinguish the sequence of various pictures of our physical reality as
objects shift in position. Sequence also allows us to distinguish between
cause and effect. The time dimension must then only be sufficient to dis-
tinguish the changes in position of the objects around us. Pearson argued
that space and time are merely modes of perception, and physics the sci-
ence that observes and governs these various modes.
These observations, which Pearson included in his Grammar of Science,
were most influential on a young Albert Einstein. In 1901, Einstein,
unable to secure an academic position because of his unconventionality,
was struggling to find work in Bern, Switzerland, as a tutor in mathematics
and physics. In 1901, Einstein placed an advertisement in the newspaper
offering his tutoring services in his apartment. A young Maurice Solovine
(18751958), a budding Romanian mathematics and philosophy stu-
dent, responded to the posting. Quickly, though, Einstein dispelled the
notion of tutoring in physics, when he uttered: It is not necessary to give
you lessons in physics, the discussion about the problems which we face
in physics today is much more interesting; simply come to me when you
wish, I am pleased to be able to talk to you.1
Instead, the pair began to indulge discussion of metaphysical issues
with Solovine. Solovine suggested that they form a group to discuss the
works of great philosophers of science. Soon, mathematician Conrad
Habicht (18761958) joined what soon became known as the Akademie
Olympia, or Olympia Academy.
These meetings in Einsteins apartment occurred from 1902 to 1904,
and also included Paul Habicht (18841948), the brother of Conrad
Habicht, Einsteins friend, the mechanical engineer Michele Besso
(18731955), his classmate Marcel Grossmann (18781936), the elec-
trical engineer Lucien Chavan (18681942), and Mileva Mari (19
December 18754 August 1948), a brilliant Serbian mathematics stu-
dent and Einstein's first wife, whom he married on 6 January 1903.
10 The Later Life andLegacy ofKarl Pearson 107
Note
1. https://en.wikipedia.org/wiki/Maurice_Solovine, accessed 3 February 2016.
Part 3
The Formation of Modern Statistics
The innovation of the first social scientists, initiated by Quetelet and fur-
thered by Galton and Pearson, was certainly evolutionary, if not necessar-
ily revolutionary. Each suggested concepts and some helpful measures in
characterizing data, but none practiced the style of science increasingly
demanded in the rapidly expanding scientific age. Rather, they were prac-
titioners and engineers of issues often relevant to social science.
While these early intellects became increasingly adept at describing
the properties of basic statistical estimators, they had not yet struck upon
the need to establish the larger question of statistical validity for an entire
model. It would take a fresh and much more rigorous geometric approach
to provide the foundation for a new science of statistics. This scaffolding
was created by Ronald Aylmer Fisher on one side of the Atlantic, and
furthered and promoted by Harold Hotelling on the other side.
11
The Early Life ofRonald Aylmer Fisher
While John may have partially sacrificed the respect a medical career
could afford, he nonetheless benefited from other trappings of an
increasingly successful family. His father purchased on his behalf a life-
long Governorship at nearby Christs Hospital, and his position within
the Piccadilly business community afforded him the opportunity to
meet and then marry Emma Mortimer (1827), a daughter of Thomas
Jackson Mortimer (17811833) and Elizabeth Mavor (ne Elsworth)
(17651816).
Thomas Jackson Mortimer maintained and furthered a tradition of
gunmaking in his family that had dated back to 1753. In this respect,
Ronald Fisher shared a family foundation with Francis Galton. The
paternal great-grandfather of Galton and the maternal great-grandfather
of Fisher both ran gun manufacturing businesses.
The Mortimer shop at 34 St. James Street had been producing top-
quality guns. After his death in 1833, his wife Elizabeth and son Thomas
Elsworth Mortimer maintained the business for a couple of years in
London, but Thomas soon moved the business to Edinburgh. By 1836,
the family firm was recognized Gunmaker to His Majesty, King William
the IV.Fifteen years later, one of their guns was awarded the Prize Medal
at Londons Great Exhibition in 1851. The firm eventually was trans-
formed into Mortimer and Sons under the leadership of Thomas Jackson
Mortimers grandson, Thomas Alfred Clark Mortimer (Fig. 11.1).
The daughter of Thomas Jackson Mortimer, Emma, and her husband
from the neighborhood, John Fisher, appreciated education and had a
sufficiently comfortable life to afford the same for their children. John
and Emma lived above the growing poulterers shop in St. James and
assisted in the family business until the untimely death of George in
1855, following a slip and fall on an orange peel.1 John chose to pass the
business on to his younger brother and, at the age of 40, turned to the
life of a leisure gentleman of London. For the next 50 years, he reigned as
the patriarch of the family, first in London, until the age of 74, and then
in retirement in Norfolk until his death at the age of 90.
Before his retirement, though, John and Emma raised all but one of
their 13 children in the apartment above the poulterers shop. Of these
children, only five survived childhood. The eldest surviving son, George
Fisher, was born on 10 August 1843, while the youngest son, John Fisher,
11 The Early Life ofRonald Aylmer Fisher 113
Edward Fisher
b: 1790 in England
m:
John Fisher
d:
b: 23 Jun 1816 in Thames Ditton,
Surrey, England
m:
Lucy
d: 1907
b:
d:
George Fisher
b: 10 Aug 1843
m:
Thomas Jackson Mortimer
d: 1920
b: 20 Nov 1781 in Wednesbury,
m: 26 Oct 1806 in Saint Bride
Emma Mortimer
d: 12 Dec 1833 in London, England
b: 13 May 1827 in Middlesex,
England.
d: England
Elizabeth Mavor Elsworth
b: 1765 in Chard, Somerset,
England
Sir Ronald Aylmer Fisher D.Sc d: 10 Mar 1816 in Chard, Somerset,
and F.R.S. England
b: 17 Feb 1890 in East Finchley,
m: 01 Apr 1917 in Upper Samuel Heath
d: 29 Jul 1962 in South Australia, b: Mar 1788 in Shaftesbury, Dorset,
England
Samuel Heath
m:
b: Aug 1818 in Finsbury, St Luke, d:
Middlesex, England
m: 08 Aug 1849 in St Mary,
MARY
Islington, England
d: 1900 in England b: 1786 in CAVERSWALL
STAFFORDSHIRE
d:
Katie Heath
b: 09 Dec 1856 in England
d: 21 Jun 1904 in Dartmouth,
Richard Herring Worth
Devon, England; Age: 49
b: 1795 in Plympton St. Mary
m: 11 Feb 1819 in Plymouth,
Elizabeth Worth
d: Jul 1838 in Stepney, London
b: 04 Mar 1824 in City of London,
Middlesex, England
d: 1900
Mary Elizabeth Alger
b: Abt. 1798 in Ashburton, Devon,
England
d: Jul 1878 in Stratton, Cornwall
was born 15 years later, on 20 May 1858. John Jr. was afforded the career
that was pre-empted for his father. He completed his medical education
at Kings College School and the associated St. Georges Hospital, where
he rose to House Surgeon. John Jr. died on 6 November 1918.
Meanwhile, while George maintained the merchant tradition, he did
so not by selling poultry. Rather, he established himself as a trader in fine
arts in partnership within the firm Robinson and Fisher. George Fishers
lan and sophistication afforded the firm a reputation as one of Londons
top fine arts auction houses.
114 The Econometricians
February 1890, the last child, a boy they named Ronald Aylmer Fisher,
was born.
In fact, Ronalds birth was a bit of a miracle. Katie had a difficult preg-
nancy followed by the heartbreaking birth of a stillborn boy. Moments
after this sad announcement, a second, smaller baby arrived. Ronald
began life as a surprise gift, and much younger than his three sisters, who
were all teenagers or young adults even before Ronald was old enough to
begin school. Ronald, more than others, also benefited from the increas-
ing comfort afforded the family.
Just as had Francis Galton, Ronald was doted upon by his older sisters,
and was instructed in the classics, learned to read at an early age and
found a fascination with mathematics even by the age of three. But, while
he was a precocious youngster, his very poor eyesight made it difficult for
him to read. Instead, his nannies, sisters and mother would read to him,
often on the subject of astronomy and geometry, something he could pic-
ture and imagine in his head. From this early introduction, he garnered
a lifelong interest in astronomy and in the mathematical reasoning that
was the basis of geometry, and a yearning for the establishment of the
new science of statistics. Even Ronalds poor eyesight became an asset. It
forced him to develop as a visual learner. He became adept at translating
mathematical concepts into geometric shapes in his head.
When Ronalds brother Alwyn, three years his senior, began to attend
day school in Hampstead Village, Ronald soon began to tagalong. While
he was much younger than the other pupils, his intellectual precocious-
ness and his physical neediness seemed universally endearing. He earned
special attention from Headmaster Greville, and his report card soon
reflected his brilliance. Despite his youthfulness by two or three years
compared to his classmates, he soon scored consistently at the top of his
class in science and mathematics.
By the age of 14, Ron was on his way to Harrow, one of the elite
preparatory colleges in England. His transition into independence and
adulthood arose for yet another reason that year. He lost his beloved
mother, Katie, at the early age of 49, following her quickly progressing
bout of peritonitis. Gone was his comfortable life on Heath Hill, his
doting mother, nurses and sisters, and his ability to excel in small classes.
He was thrust into an elite school with older boys, many of whom were
116 The Econometricians
brilliant themselves. Yet, despite these hardships and his almost debili-
tatingly poor eyesight, he earned the schools highest prize, the Neeld
Medal, after only two years at the school. He excelled with the specialized
tutoring students at Harrow received, but, because tutoring was by lamp-
light in the evening, he absorbed subjects using geometric insights he
could picture in his head rather than on dimly lit pages that challenged
his poor eyesight. It was there that he demonstrated his uncanny ability
to picture the solution to complex problems.
Fortunately, Ronalds accomplishments secured him scholarships. Less
than two years following the death of his mother, his father was bank-
rupt, and the family could no longer afford Harrow. At the age of 16,
Ronald needed to rely on his innate intelligence by securing a series of
scholarships. Then, in October of 1909, at the age of 19, Ron was off to
Cambridge with a full scholarship.
At Cambridge, Ron was popular among his fellow students, but per-
haps not for the slightly disheveled nature of his clothes, or his somewhat
odd body shape. He was nicknamed Piggy, but seemed unoffended. His
classmates were astounded especially by his mathematical acumen, and
his ability to make up for months of lack of attention with hours or days
of intense study. He also demonstrated a strong interest in the new field
of genetics in his four-year stay at Cambridge.
This was a period of infancy in the study of genetics, pioneered only
four decades earlier by the work of the Augustinian friar Gregor Mendel
(20 July 18226 January 1884) and his study of the inheritance of genetic
characteristics in peas. Also not fully appreciated yet was the notion of
natural selection as pioneered by Francis Galtons cousin Charles Darwin.
But while Darwins theory remained debatable, Ronald Fisher saw the
profound link between his work, Mendels genetics, and the growing
influence of eugenics as championed by Galton.
By the time Fisher entered Cambridge, the new statistical journal
Biometrika Karl Pearson had founded was seven years old, and Galtons
legacy had been cemented with the dedication of the Galton Laboratory
at the University of London, headed by Karl Pearson. Fisher was well
subscribed to Galtons increasingly populist belief that humanity can be
improved through science. Statistics was the methodology that would test
and guide the success of the study of eugenics he formulated to improve
the genetic stock of humanity.
11 The Early Life ofRonald Aylmer Fisher 117
expected him to invest more in his wardrobe than he could afford so that
he may look a banker instead of the threads of a shabby statistician.
With the onset of the Great War in 1914, Fisher tried to volunteer for
active duty, but was rejected because of his extremely poor eyesight. He
tried his hand on occasion at teaching at preparatory colleges, but he did
not have a natural teachers empathy for his students, despite his obvious
academic brilliance. Meanwhile, to hedge his vocational bets, he simul-
taneously tried his hand at hobby farming. There he raised pigs and pur-
sued their horticulture with the keen eye of a statistician. He considered
as unsuccessful his efforts as a school teacher, but with his college chums
off at war, and his own academic success thwarted at each turn, he had
turned to farming and teaching to pay the bills.
The interest in eugenics, and his experiences working on the Canadian
farm, made Fisher interested in starting a farm of his own. Through his
college friends, he had come across a woman named Gudruna Guiness.
The granddaughter of Henry Grattan Guinness (11 August 183521 June
1910), a famous preacher of the Ulster Revival of 1859, and the great-
great-granddaughter of Arthur Guinness, the founder of the Guinness
brewery, Gudruna was a spirited young woman married to one of his col-
lege friends. Gudruna entertained Rons animal husbandry interests, and
offered her emotional support for his interest. They pursued this part-
time interest together, even as Gudruna fanned Rons interest in find-
ing a companion. She arranged to have him meet her 16-year-old sister,
Ruth Eileen Guinness (15 April 190015 January 1982), with whom the
much older Ron struck a romantic interest. Knowing the age difference
would be controversial, Ron and Eileen delayed their marriage until she
turned 17 years old.
Eileens father had died in 1915, and Eileen managed to hide from
her family her interest in Ron, and his interest in her, for another two
years. Just 11 days after her 17th birthday, Ron and Eileen married, near
Rons college work in Kent. They left that day to visit with Rons family
in Streatham, then moved to Bradfield to occupy a cottage, raise pigs and
produce, and permit Ron to teach at nearby Bradfield College.
Together, Eileen and Ron had a boy, named after Rons father, and
then, in 1920, as Rons family, and his sister-in-law moved to adjoining
cottages in the village of Markyate, and Ron and Eileen then had a girl,
120 The Econometricians
named after Rons mother. They went on to have another boy, and six
girls together, one of whom died as a young child.
Meanwhile, Fisher had come to admire and respect the son of Charles
Darwin, Major Leonard Darwin (15 January 185026 March 1943), and
considered the Major a mentor. Eventually, as president of the Eugenics
Society, Darwin was able to offer Fisher a modest stipend to perform
statistical work for the Society. This afforded Fisher some time to devote
to his statistical interests, which parlayed into a paper on the use of the
correlation coefficient. However, when an academic quarrel erupted
between Darwin and Pearson, Fisher came to the defense of his mentor,
and somewhat alienated himself from Pearson, who could be prickly with
those whom he perceived as detractors. Unfortunately, this feud would
accelerate and come to define Fishers career.
Note
1. Box, Joan Fisher, R.A.Fisher The Life of a Scientist, John Wiley Publisher,
NewYork, 1978, at p.5.
12
The Times ofRonald Aylmer Fisher
Pearson had recognized that the data observed within the biologi-
cal sphere failed to follow the familiar normal distribution commonly
observed in physics and assumed by early social and biological staisti-
cians such as Quetelet and Galton. Pearson sought to parameterize these
data distributions using his frequency curves, not only with the first and
second moments of mean and variance, but also with weightings of third
and fourth moments, as convenient. In the process, statistics had tempo-
rarily been diverted from more fully describing and deriving the proper-
ties of the normal distribution. By returning to the underpinnings of the
normal distribution, under the Laplacian and Gaussian observations of
the central limit theorem, Fisher was shoring up the foundation of mod-
ern statistics. But, clearly his approach was a dramatic departure from
Pearsons methodology.
Fisher was most clear on his concern that Pearsons description of dis-
tributions through various weighted moments may offer some simplicity,
but was entirely arbitrary. Gauss had derived his distribution based on
the optimization of a likelihood function. As had Gauss, Fisher began to
champion the maximum likelihood approach.
Fishers first volley into the Fisher-Pearson fray occurred when he
was only 22 years old. His approach was influenced by a paper pub-
lished in 1908 under the pseudonym Student. At that time, a chemist
named William Sealy Gossett was an employee of the firm Rons wifes
great-great-grandfather had started, the Guinness Brewing Company
of Dublin, Ireland. The company had instructed Gosset to analyze
farm productivity data for plots associated with the brewery. Gosset
had brought to the attention of his mentor, F.J.M. Stratton, some
results on the distribution of observations. In turn, Professor Stratton
brought these results to his young student, Fisher. In 1912, Fisher con-
tacted Gosset and showed Gossett that the results do not precisely fit
the normal distribution, but can be expressed as a modification of the
normal distribution that takes into account the finite sample size, as
opposed to the large, or infinitely sized, samples required by the central
limit theorem.
Fishers correction of Gossets Student t-distribution was relatively
minor, but it removed the ad hoc nature of Gossets proposal by provid-
ing a justification for the recasting. Gosset confided in his friend, Pearson,
124 The Econometricians
whom he had come to know when he visited Pearsons lab for the aca-
demic year 190607 that Fisher had reworked his results. In a letter to
Pearson, he asked Pearson what he thought about Fishers correction.
Had Pearson fully comprehended and accepted the note from a
22-year-old graduate student in 1912, he might have offered a correction
in his journal to Gossets original publication of a paper on the Student
t-distribution. Pearson never published such a correction. Instead, Fisher
separately published On an Absolute Criterion for Fitting Frequency
Curves,2 in which he derived a maximum likelihood estimate of variance
when the sample size is small. In doing so, he constructed a now familiar
standard deviation estimate for a sample of size n that includes the now
equally common concept of degrees of freedom.
( x m)
2
s.d. ( x ) = .
n 1
189521 June 1980), had taken over the lab following his father retire-
ment, and had published an article in Biometrika that generated results
dramatically contrary to his fathers prediction, and almost precisely what
Fisher had predicted. While few statisticians had the level of understand-
ing of either Egon Pearson or Fisher, and still fewer would take Fishers
side over Pearson, father or son, even if the evidence grossly favored
Fishers analyses, almost nobody wanted to enter the fray. Fishers vindi-
cation would be delayed for years.
While Fisher would not realize the success he deserved for many years,
he continued to work to recast modern statistics on a foundation as rig-
orous as would Gauss. In 1922, Fisher published the single most sig-
nificant statistical work since Gauss 1809 magnum opus. His On the
Mathematical Foundations of Theoretical Statistics established the vocabu-
lary of statistics employed ever since, and firmly cast its methodology
based on the principle of maximum likelihood. In this, and a subsequent
addition a couple of years later, Fisher described the goal of statistical
measuresto provide statistics that are as efficient as possible in sum-
marizing all available information. Based on that criteria, it was evident
that the measures he developed or improved were the standard to be
employed in all statistics going forward.
In addition to the cogent statements of various familiar measures and
his development of the new analysis of variance, Fisher also described the
now familiar F-test. Yet, as productive as he was, Fisher was producing
these great works not from a comfortable scholarly position aloft in the
ivory tower, but as a working statistician at an agricultural field station.
His work at Rothamsted was indeed his most significant.
Rothamsted was by no means the University of Londons Galton
Laboratory, but neither was it a rural backwater. It was there that Ronald
Fisher supervised a few assistants and budding statisticians in the orga-
nization and study of a massive amount of collected crop data, on the
surface, but also on the development of modern statistical theory. His
major treatise, published in 1922,4 and its extension in 1924 were a
tribute to Rothamsted and created the foundation for modern statistics.
And it was there that many notable statisticians followed in his found-
ing footsteps.
128 The Econometricians
Notes
1. Fisher, R.A., The Correlation between Relatives on the Supposition of
Mendelian Inheritance, Philosophical Transactions of the Royal Society of
Edinburgh, Vol. 52, 1918, pp.399433.
2. Fisher, R.A., On an Absolute Criterion for Fitting Frequency Curves,
Statistical Science, Vol. 12, No. 1, February, 1997, pp.3941.
3. Student, Probable Error of a Correlation Coefficient, Biometrika, Vol. 6
No. 2/3, Sept. 1908, pp.302310.
4. Fisher, R.A., On the mathematical foundation of theoretical statistics,
Philosophical Transactions of the Royal Society, 1922, pp.309368.
13
Ronald Fishers Great Idea
While Pearson had spent a lifetime formulating statistics for such indi-
vidual predictors as a distributions mean and variance, and Fisher helped
improve Pearsons statistics, Fisher also offered a new and groundbreaking
insight into the overall power of statistical models. While at Rothamsted,
Fisher also refined a geometric interpretation of statistical measures that
dated back to Gauss, and integrated most of the statistical measures we
employ today.
To understand an insight Ronald Fisher originally had when he was
just a 19 years old, let us harken back to why Gauss referred to his
Gaussian distribution as the normal distribution. To a physicist or geo-
metrically trained mathematician, the term normal is synonymous with
orthogonality, or the angle 90. It is the direction from a given ray, plane
or hyperplane along which a vector can minimize the distance between a
given point off the plane to a point on the plane. This direction normal to
the plane is thus one of great geometric significance. This normal ray was
both the key to minimizing the sum of squared errors (SSE), or, likewise,
the square root of the sum of the deviation distance between an observa-
tion subject to measurement error, and the path which it is expected to
follow. This distance then represents the minimum of the root of the SSE.
with the same length as Y would then create the same SSE.Such rays would
sweep a sphere with a radius equal to the square root of y12 + y22 + y32 .
Then, the accuracy of the alternative hypothesis that the underlying
expression is given by X is given by the chance that the observed ray Y
would fall as close to X relative to any such ray of the same length as Y.
This relative probability is given by the area swept out by rays of length Y
within the same distance between Y and X relative to all possible rays of
length Y, or the surface area of a conic slice around X as a share of the
entire surface of a sphere of radius X. This share can then be interpreted
as a test of the significance of .
Geometrically, this measure is proportional to the angle between the
ray Y and the ray X. The test of significance then compares the area of
the predicted cone compared to the entire area of the sphere. The value
of can then be stated based on this ratio of confidence. Fishers t-stat
performs that calculation of the relative areas.
We can also use Pythagoras theorem to analyze variances using this
same methodology. The square of the total length of the observational
vector, y12 + y22 + y32 , can be broken into two components, the regres-
sion sum of squares plus the SSE. Mathematically, the domain of the
observational vector is given by its dimensions that can vary, that is, the
number n individual observations. Then, the mean sum of squares per
132 The Econometricians
SR2 / nR
F= = ( nE / nR ) cot 2 ( ) ,
SE2 / nE
where the cotangent of the angle between the observational and the
regression vectors gives the ratio of the adjacent regression ray and the
opposite error ray. Fisher proposed that a transformation of this F-ratio,
given by what he labeled the z-test, as:
z = (1 / 2 ) log F ,
r=
xy ,
x y
2 2
which, Fisher observed, is given by cos(). At each turn, the relevant sta-
tistics had been reduced to a firm geometric interpretation, just as Fisher
had learned to imagine since he was a child.
While at the Rothamsted Experimental Station, Fisher published his
results in his Statistical Methods for Research Workers, which he wrote over
two years beginning in 1922. His brilliant interpretation of modern sta-
tistics was not particularly well received in Britain, but it was embraced
in the USA, which was less allied with the influential Pearsons and the
journal Biometrika. More theoretical statisticians had remained wed to
the arbitrary multiple moments methodology of the statistical giant, and
editor of Biometrika. Meanwhile, young practitioners at first found the
theoretical foundations established by Fisher confusing and unnecessary.
His foundational approach nonetheless slowly gained appreciation and
academic traction. It is now universally appreciated for its rigor, inter-
pretation and significance in the development of measures of obvious
geometrical significance.
mi = p * ai, + q * bi, + r * ci ,
134 The Econometricians
vi = p * ai, + q * bi , +r * ci mi .
n
min ( pai + qbi + dci mi ) .
2
p , q ,r
i =1
Gauss used various methods to solve for these systems of linear equa-
tions in his solutions to such problems as the reappearance of an asteroid.
Many mathematicians also developed favorite methods. However, the use
of matrices was not developed until Arthur Cayley (16 August 182126
January 1895). Cayley had defined the method of matrices in a German
journal article written in French in 1855.1 We now recognize the system
of equations and can solve the system with relative ease. In 1932, two
mathematicians employed Cayleys method to demonstrate how it might
be used as a notation to simplify the traditional linear regression solu-
tion discussed earlier. Herbert Westren Turnbull (31 August 18854 May
1961) and Alexander Craig Aitken (1 April 18953 November 1967)
produced a version of the now familiar expression for the solution of the
parameter matrix:2
= ( XX ) X y.
1
13 Ronald Fishers Great Idea 135
Fisher had derived close relatives to this now familiar form, but
without the much more compact matrix notation. Nonetheless, it was
Fisher who took this formulation of the linear regression model and
derived the various validity tests we use now in his Statistical Methods
handbook. These include the F-tests and t-tests, analysis of variance and
the R2 value.
A modern approach to Fishers results is described next. The method
of least squares as applied to the linear regression model has two compo-
nents. First, we show that the method provides an unbiased set of estima-
tors for a line that best fits a set of observed data. Second, we show that
this method produces its unbiased estimate of the linear parameters so
long as the errors are symmetric, with a mean of zero, and each obser-
vational error is independent of the next. They need not follow a more
formal distribution function.
Let us consider a vector of observation points y that linearly depend
on a set of independent variables X, which include the constant 1. Let
there be a vector of linear coefficients b such that, in the absence of any
observational errors,
y = Xb.
y i X i b.
SSE = ( y Xb )( y Xb ) .
Gauss determined the expression for the linear coefficients and demon-
strated that these coefficient estimates minimize the SSE.To see this, let
136 The Econometricians
us differentiate the SSE with respect to the estimated coefficients and set
these terms to zero:
dS dS
( y Xb) ( y Xb)
0= SSE =
db db
dS
= ( yy bX y yXb + bX Xb) = 2 X y + 2 X X ,
db
= ( XX ) X y.
1
Let us now show that this estimate is unbiased. Let the vector of errors
be given by , with each error i assumed to have a mean of zero and a
variance i2 . Since these error terms are independent, the covariances
ij=0. Then, we can calculate the expected value of our estimates of the
coefficients:
( )
E = E(( X X ) X (( X + )
1
(
= + E ( X X ) X
1
)
= + ( X X ) X E ( )
1
= .
Next, we can discover the variance of the estimates for the coefficients.
These are given by:
( ) (
Var = E ) ( ) = E (( X X ) )( X X ) ))
1
X 1
X
(( )
= E X X )1 X X ( X X )
1
) = E (( X X ) X ) X ( X X ) )
1 2 1
13 Ronald Fishers Great Idea 137
(
= 2 E ( X X ) X X ( X X )
1 1
) = 2
( X X )
1
.
Notes
1. Cayley, Arthur, Remarques sur la notation des fonctions albebrai-
ques, Crelles Journal fur riene und angewandte Mathematik, Vol., L,
1855, pp.2820285.
2. Turnbull, H.W. and A.C. Aitken, An Introduction to the Theory of
Canonical Matrices, Blackie and Sons Ltd., London, 1932.
14
Later Life andLegacy ofRonald Fisher
There are but a few scholars in any discipline who are simultaneously
claimed by multiple disciplines as their own. The Great Mind John von
Neumann is one such scholar. The major award in computer science is a
tribute to his leadership and scholarship in the development of computers
and their programming. Physicists remember him for his contributions,
and the Manhattan Project relied on his expertise in chemical engineer-
ing. He was a disciple of David Hilbert and an extraordinary mathema-
ticians, and he was perhaps the most significant father of game theory
that added so much to so many disciplines. Finally, he is recognized as
one of the cleverest economists who also made significant contributions
to finance theory. Ronald Fisher shares with von Neumann this unique
characteristic of recognition by scholars from many disciplines.
Fisher remained a giant among colleagues at Rothamsted. He was cer-
tainly the brightest to have ever joined the illustrious institution, even
if many bright scholars followed in his footsteps there. None, however,
likely exceeded his degree of eccentricity, nor his ability to scare and
intimidate colleagues at one moment, and then smirk and joke with
them the next. He had more freedom at the Rothamsted Institute than
most in academics, and was only required to lend his incredible skill
These were battles Fisher took to his grave. Neyman delicately removed
himself from the almost constant din of professional jealousies when he
moved to the University of California at Berkley shortly before World
War II.Nonetheless, Fisher remained embittered because of the margin-
alization he experienced.
Regardless, Fisher left the statistical and finance world with a series
of innovations beyond his elaboration and geometric justification of
Students t. We now regularly employ his chi-squared and F-tests, and
practicing statisticians use his definitions, from the meaning of probabil-
ity to the concepts of efficiency and consistency. Fisher also originated the
idea of hypothesis testing, and even recommended the adoption of the
95% confidence level that has become the de facto standard for much
hypothesis testing ever since. Even the F-test, which measures the degree
to which the variance from two populations are equal, originated with
Fisher. Yet, it was the Iowa State University statistician and mathema-
tician George Waddel Snedecor (20 October 188115 February 1974)
who actually formulated the appropriate distribution to describe the
extent to which the ratio of two variances approach unity. He named this
ratio of chi-square the F-test in honor of the brilliant Fisher who spent
time in Ames, Iowa, during one of his most productive and influential
visits to the USA.Actually, Fisher preferred other methods to compare
and analyze variances, and did not readily adopt the measure named in
his honor.
Nonetheless, Snedecor was instrumental in spreading the popularity
of Fisherian statistics in the USA.He wrote his own guide for research-
ers, his 1937 text Statistical Methods, that covered the same materials
as had Fishers Statistical Methods for Research Workers, but without the
cumbersome rhetoric, and with more examples that could guide practical
statisticians. Snedecors book sold hundreds of thousands of copies and
remained one of the most cited statistical treatises for decades following,
even though it added little if any to Fishers intuitions. In fact, Snedecor
quite wittingly became Fishers best ambassador in the world of statistics
and among US statisticians. His accessible approach was also instrumen-
tal in the formation of many departments of statistics of the type Fisher
had long advocated.
14 Later Life andLegacy ofRonald Fisher 143
Beyond his relationship with Egon Pearson and Jerzy Neyman, and
while Fisher constantly quarreled with colleagues and administration
alike, he was universally appreciated by the students and assistants in
his laboratory. They found him intellectually generous, just as had his
colleagues at the experimental station. He fed young researchers ideas and
assisted them with their analyses. He was paternal, and often jocular, and
not particularly overbearing, despite a brilliance obvious to everybody.
He would even include them in his active social calendar conducted at
his home over weekends.
Fisher also enjoyed the important role as a eugenicist. He was the edi-
tor of the Annals of Eugenics, an academic trapping associated with his
Galton professorship. From that perch, he was able to keep his pulse on
the development of eugenics and statistical biology. But, while Pearson
used his position as editor of Biometrika to publish articles of his liking,
and submit to rigorous refereeing and delay those he did not, Fisher,
perhaps because he was too often subject to the arbitrariness of referees
unsympathetic to his novel approach, used the editorship as an opportu-
nity to expedite the publication of research. For that he was appreciated
by his colleagues and scholars.
The journal also afforded Fisher and his students an avenue to publish
their own work. Quickly, the Annals became a most progressive, respected
and highly regarded journal, with Fisher at the helm. In doing so, Fisher
helped forge both modern statistics and modern genetics.
Later in his career, Fisher became more intensely interested in issues
of biology. He was a genius in experimental design, and played an influ-
ential role in the resolution of many important agricultural and medical
issues of his day, from a statistical biological perspective. He was com-
missioned by the United States Department of Agriculture to provide
knowledge transfer to some large land-grant American public universi-
ties so that they may better enhance their statistical expertise and agri-
cultural effectiveness. In essence, these emerging statistical departments
were constructed in his mold. His reputation was also extending world-
wide through the late 1930s and the 1940s, throughout North America,
Europe, India and Australia.
Fisher and his Galton Laboratory also increasingly used their expertise
to assist in the rapid development of blood testing and serology that is
144 The Econometricians
named after him. And, his paper Gene Frequencies in a Cline Determined
by Selection and Diffusion was the first to use a computer to calculate a
solution in a biology paper.2
Fisher was well regarded late in life. He earned the Royal Medal from
the Royal Society in 1938, its Darwin Medal in 1948 and the Copley
Medal in 1955. The award was:
Notes
1. Fisher, R.A., The Genetical Theory of Natural Selection, Clarendon Press,
Oxford, 1930.
2. Fisher, R.A., Gene Frequencies in a Cline Determined by Selection and
Diffusion, Biometrics, Vol. 6, No. 4 (Dec., 1950), pp.353361.
15
The Early Life ofHarold Hotelling
While the names Gauss, Galton and Fisher may not be well known to
finance students of statistics, the name Hotelling often elicits at least
some acknowledgment. This is perhaps because he helped establish sta-
tistics, economics and econometrics from a distinctly American perspec-
tive. Fisher was experiencing resistance in England over his theories that
defied the powers that be in Britains statistical community. America was
more receptive. Its embracement of Fisherian statistics caused the center
of mass of the statistics world to migrate across the ocean. The primary
proponent was an American mathematician and economist whose ances-
tors made the same migration as founders of Americas first colonies.
The mother of Harold Hotelling, Lucy Amelia Rawson, was an off-
spring of one of the first Secretaries of the Massachusetts Bay Colony.
Edward Rawson (16 April 161527 August 1693) was born in the vil-
lage of Gillingham, Dorsetshire, England. His namesake, grandfather
Edward Rawson, was a wealthy silk and woolen merchant of Colnbrook,
west of London. When the grandfather died in 1603, he left two young
sons, Henry and David, to run the family business. David became a tai-
lor in London, and married Margaret Wilson, the daughter of Reverend
William Wilson of Windsor. David died when their three children were
still very young. Their young son Edward was only two years old at the
time. His mother also died 11 years later, which left Edward in the care
of his extended family.
The Rawsons and the Wilsons lived comfortably, and young Edward was
hence well educated. When he married Rachel Perne, the Wilson and Perne
families were already establishing religious links to the Colonies. Edwards
wife, Rachel, was the daughter of Thomas Perne, and the niece of one of the
first settlers in Cambridge, Massachusetts, the Reverend Thomas Hooker.
Edwards maternal grandfather, the Reverend William Wilson of Merton
College, Oxford, had rose to Canon of St. Georges Chapel in Windsor.
Edwards uncle, Edmond Wilson, was a wealthy physician in London who
had donated 1000 pounds toward establishment of a Massachusetts Bay
Colony in 1633. Another uncle, Reverend John Wilson, immigrated to the
Colony and ministered over the First Church of Boston.
The young couple Edward and Rachel married in 1636, had a daughter
who they left with family in England and migrated to the Massachusetts
Bay. By 1637 they were settled in Newbury. At the remarkably young
age of 23, Edward was selected to be the Public Notary and Register for
Newbury, and was continually reelected to that position until 1647. His
willingness to accept increasing responsibilities led to his appointment as
the Secretary of the Massachusetts Bay Colony, a position which he held
for another 36 years (Fig. 15.1).
Together, Edward and Rachel Rawson had 12 children, of which
only 9 survived past childhood. One of their children, Rebecca Lawson
(23 May 1696?) was the heroine of the popular 1849 book by John
G.Whittier entitled Leaves from Margaret Smiths Journal in the Province
of Massachusetts Bay. While three of their sons eventually returned to
England, including Edward following his graduation from Harvard
College in 1653, two sons, William and the Reverend Grindal (23
January 1659), remained in the New World.
Born in Boston, William Rawson (21 May 165120 September 1726)
moved south, first toward Dorchester, and then to Braintree with his
wife Ann Glover (165329 July 1730). Together they had a son, William
Rawson (8 December 1682October 1769), who married Sarah Crosby
(27 July 16841734) on 26 October 1710, and settled in Mendon,
15 The Early Life ofHarold Hotelling 151
Alice Blower
Pern Rawson b: 30 Jun 1615 in Sudbury, Suffolk,
b: 24 Feb 1741 in Mendon, England
Worcester, Massachusetts, d: 03 Nov 1690 in Braintree,
United States Norfolk, Massachusetts, USA; Y
m: 04 Feb 1762 in Mendon, Thomas Cook
Worcester, Massachusetts,
United States b: 1689 in Mendon, Worcester,
d: Massachusetts, United States
m:
d:
Margaret Cook
b: 18 Dec 1711 in Mendon,
Worcester, Massachusetts,
United States
d:
Mary Carey
b: 1693 in Mendon, Worcester,
Massachusetts, United States
d:
Thomas Rawson
b: 04 Dec 1776 in Mendon,
Worcester, Massachusetts,
United States
m: 26 Oct 1804 in Attleboro, Bristol,
Massachusetts
d: 06 Nov 1866 in Cumberland,
Providence, Rhode Island,
United States; Y
York
d: 11 Mar 1857 ; Age: 66
Annatie Osterhout
b: 1740
d: 1793 ; Y
153
During this era of the early 1600s, the Dutch West India Company
was based on Amsterdam at the heart of one of the worlds wealthiest
nations, fueled by rapidly expanding world trade. The Dutch political
model was unlike some of their competing empires. They did not oppress
religion, and they relied on private enterprise as the vehicle for their colo-
nial ambitions.
The Dutch East India Company had sponsored the English explorer
Henry Hudson (c.15651611) to explore an easterly passage to Asia. He
was instructed to find a passage along the coast of Russia, which is now
called the Northern Passage. When his ship Halve Maen ran into ice,
he discarded his orders and instead turned his ship west to seek a route
through North America.
By 2 July 1609, Hudson had reached the Grand Banks, south of
Newfoundland, and had passed Nova Scotia a couple of weeks later. Over
the next month, the ship had navigated past Cape Cod, Massachusetts, and
traveled further south to the Chesapeake Bay. From there, Hudson turned
back north, past the Delaware River, and, on 11 September, began to sail up
the Hudson River adjacent to NewYork City. Before the end of September,
he had reached Albany, the present-day capital of NewYork State.
Hudson then returned to Europe. He first arrived in Dartmouth,
England, where he was arrested as authorities sought his logbook. Instead,
he passed the log to the Dutch ambassador so that the Dutch could lay
claim to his discoveries in the New World. The region from NewYork
City to Albany was subsequently settled by the Dutch as a commercial
avenue for the newly formed Dutch West India Company. The com-
panys goal was to establish farmland along the Hudson delta region to
feed their soldiers and workers. Nieuw Amsterdam was established after
its purchase from the indigent population for 60 guilders.
The Company sought settlers, who would be offered land, and patrons
who would earn trading privileges. A mix of the wealthy settlers and
cheap labor were sent to settle the region between NewYork and Albany.
Among them was Mathys Coenratsen Houghtaling (16441706), who,
in 1655, was an 11-year-old orphan living in an almshouse in Amsterdam.
He was sent to the New World in that year, eventually briefly settled first
in Coxsackie and then in the farming community of nearby Kinderhook,
where he shared a farm with his father-in-law, Hendrik Marseli. By the
15 The Early Life ofHarold Hotelling 155
late 1600s, he was established on his own farm, with his wife Maria
Hendrikse (17 March 16471706).
Their firstborn in marriage, Conrad Mathys Houghtaling (16671745),
married Tryntje Willemse Van Slyck (166797) in 1685in Albany, and
they had two sons before Tryntje died at the age of 30. Their second son,
Willem Houghtaling (17 January 16921742), remained for much of
his life in Albany, where he married Helena Uzille (27 March 16965
January 1756) on 9 November 1716. They had five children together in
Albany, including a second son Pieter (19 October 171813 November
1770), who also lived in Albany all his life and married a local woman
Annatje Becker in 1747.
The second child, and first son of Pieter and Annatje, Gerrit
Houghtaling (26 May 17511821), also lived in Albany for most of his
life, and married Annatie Osterhout (174093), with whom they shared
a son, Peter (19 January 179011 March 1857). Peter, too, remained
in the Albany County area, married Catherine LaGrange and had nine
children with her, including a first son William Helmus Houghtaling (16
February 18137 May 1887), their second child.
In the 1860s, the Hotellings, too, made their westward migration. The
son William Houghtaling moved to Iowa, with Mary Ann Yost (10 March
181630 June 1895), whom William had married on 12 March 1835,
and with seven children in tow. Their third child, and first son, Addison
Hiram Hotelling (22 March 184317 May 1896) had by then adopted
an Americanized version of the family name, and lived much of his life
in Iowa. It was there that he met Nellie Tuttle (7 August 18447 March
1929), and had five children together, including their second child, and
first son, Clair Alberto Hotelling (22 June 18695 June 1944).
The convergence of the Hotelling and the Rawson families came about
through an unlikely path, but with farming in common. With a daughter
Marjorie (19 September 1904) just born, the family set out in 1905 for
Washington State and arrived by train in Seattle, with by then five chil-
dren in tow, Clair had been a hay merchant, which was an essential fuel
for the transportation segment of the supply chain as hay fed the horses
that moved the goods at the turn of the century before the widespread
use of automobiles and trucks. Ford began selling cars that were almost
immediately popular and affordable in 1903, and Clair began to realize
that the invention of the automobile may soon doom his comfortable
hay-selling business, and decided to seek opportunity in the West.
The family were strict observers of the Methodist tradition, and the chil-
dren were raised within the Church community in Seattle. Unfortunately
for his father, though, the family lost their business and savings with the
Recession of 1907, and Harold found himself preparing for college and
supporting himself through school at the University of Washington by
working odd jobs.
One such job at a local newspaper, the Puyallup Herald, caused him to
shift his interest from electricity and science to journalism. But with the
draft of the Great War, he had to suspend his college education briefly.
Despite his college studies, he was ordered to army camp, where he was
promptly kicked by a mule, which resulted in his discharge from the
army. When he returned to college, he found the journalism department
was in disarray, which made it difficult to resume his studies. He was able
to substitute courses in economics and thus completed an economic-rich
course of study. He then went on to study mathematics in Seattle and at
Princeton and Chicago.
Hotelling was most interested in mathematics. One of his most sub-
stantial influences when he resumed graduate studies in 1920, following
his BA graduation in 1919, was his mathematics professor Eric Temple
Bell (7 February 188321 December 1960). Trained at Stanford, the
University of Washington and Columbia University, Bell become rec-
ognized in mathematical theory, in particular the Bell series in number
theory, but also in his biographies of great mathematicians. Bell was
teaching at Washington when Harold was a student, and, throughout
his career, inspired dozens of mathematicians and economists not only
158 The Econometricians
through his insights but also in his anecdotes and perspectives about the
mathematicians who preceded him, through a series of essays collected
in his work Men of Mathematics: The Lives and Achievements of the Great
Mathematicians from Zeno to Poincare.1 Bell inspired Hotellings interest
in mathematics in the year 1920.
Someone else also inspired Hotelling that year. On 20 December 1920,
Hotelling married Floy Tracy (31 December 18902 October 1932). Less
than three years later, they had a son together, Eric Bell Hotelling (12
May 19232 January 1991), named after his inspiring professor Eric Bell.
The couple had a daughter together as well, Murial (11 October 1925).
Unfortunately, neither of these children would enjoy their mothers care
and love beyond grammar school. Floy died short of her 32nd birthday.
Hotelling received his masters degree in mathematics in 1921 and had
hoped to attend Columbia with a fellowship to study for a PhD in econom-
ics, given his economics studies and his advanced mathematical training.
When his application to Columbia was rejected, he instead accepted a fellow-
ship to study additional mathematics at Princeton. There, he studied under
Oswald Veblen (24 June 188010 August 1960), the nephew of the famed
economist Thorstein Veblen (30 July 18573 Aug. 1929), and completed a
doctorate in mathematics on topology and differential geometry in 1924.
With a doctorate at hand, Hotelling found employment as a research
associate at the Food Research Institute of Stanford University. It was there
that he wrote his first two important papers, one on the mathematical
economics theory of depreciation, and another on correlation ratios aris-
ing from random data in statistics. His research interests in statistics was
piqued, and he became aware of Fishers work on the provision of rigor-
ous foundations to statistical concepts, his publication Statistical Methods
for Research Workers. He quickly became an American admirer of Ronald
Aylmer Fisher, even as Fisher struggled to gain respect in his home coun-
try of England. Hotelling had found a kindred spirit in Fisher, who also
pictured statistical concepts from a geometric perspective.
Note
1. Bell, Eric Temple, Men of Mathematics, Simon and Shuster, NewYork,
1937.
16
The Times ofHarold Hotelling
Then, with the onset of World War II, Hotellings Methodist values
were further stimulated. He advocated for a war effort to defeat Hitler
through the creation of a Statistical Research Group at Columbia that
could apply scientific principles to such problems as the maximization of
the range of aircraft and of problems associated with the weather. He also
finally convinced Columbia University to establish one of the countrys
first departments dedicated to statistics. He wrote his argument in the
form of a paper entitled The Teaching of Statistics in which he concluded:
challenge and left the academic year after the end of World War II to start
a program in Chapel Hill. There, he chaired his own new department
and was also offered the associate directorship of an Institute of Statistics.
From his position, he devoted energy once applied to his prolific research
to the administration of one of the top programs in statistics worldwide.
He remained at the University of North Carolina for the rest of his career
as the Kenan Professor of Statistics, from 1961 on, until the bestowment
of Professor Emeritus upon his retirement in 1966.
Notes
1. Hotelling, H., Review: Statistical Methods for Research Workers by
R A Fisher, J.Amer. Statist. Assoc. 22 (1927), 411412, at 412.
2. Hotelling, Harold. The generalization of students ratio. Annals of
Mathematical Statistics (Institute of Mathematical Statistics) 2 (3),
1931, pp.360378.
3. Fisher, R.A., in Nature, 17 August 1924.
4. http://www-groups.dcs.st-and.ac.uk/~history/Biographies/Hotelling.
html, accessed 25 February 2016.
17
Harold Hotellings Great Idea
( xi )
2
n
1
f ( x1 , xn : , ) = e 2 2
.
i =1 2
( xi )
2
1
ln ( f ) = nln ( 2 ) nln ( ) .
2 2 2
ln ( f ) ( xi )
= =0
2
xi
Or = .
n
Gauss demonstrated that the maximum likelihood estimate for the mean
is the mean itself. In other words, the estimate is unbiased. In addition, by
differentiating with respect to the standard deviation, he showed that:
17 Harold Hotellings Great Idea 167
ln ( f ) n ( xi )
2
= + =0
3
( xi )
2
Or = .
n
Again, the estimate of variance is an unbiased estimate of the true
statistic.
Fisher had asserted a number of asymptotic properties of the method
of maximum likelihood as the sample size grows large, to infinity. As the
sample size increases, we find the estimates are:
at the time, and perhaps because Hotelling had not been drawn into
the statistical feuds raging in London and Cambridge at the time that
divided people (mostly) into Pearsons camp, and (occasionally) into
Fishers camp, or more often neutrality, Hotellings review in 1927 of
Fishers handbook was perhaps one of the most probing and unbiased
considerations of Fishers contribution. Upon the publication of Fishers
second edition, Hotelling followed his rave review up with another that
exposed Fisher and his foundations to an even broader audience. Then,
with his visit to Rothamsted, Hotelling was able to appreciate even more
the brilliance of Fisher.
By the time Hotelling visited with Fisher, Fisher had already articulated
a theory of maximum likelihood based on his principles of consistency, effi-
ciency and asymptotic normality. Fisher was reviving a methodology that had
actually been initiated by Pearson. As you recall, Pearson was fond of describ-
ing relationships based on moments. In doing so, he postulated a likelihood
ratio, defined as the ratio of the distribution, or frequencies, of observed data
over the postulated frequencies from some sort of postulated estimation. By
changing the parameters of his estimates slightly, he could recalculate this
ratio. He employed a Taylors series expansion to explore this ratio as con-
stants are perturbed to discover the constants that yield the best fit.
Pearsons likelihood ratio approach was actually very close to the maxi-
mum likelihood technique more fully described later by Fisher, but Pearson
did not take his results far enough to realize the generality of his approach.
Early in his career, Fisher corresponded with Pearson to advocate an
approach based on the minimization of various aggregate error measure-
ments, but Pearson panned the idea and noted he had gone down such a
fruitless path years earlier. Fisher took this as a challenge to place the method
of maximum likelihood on a firm theoretical foundation. His work, if suc-
cessful, in combination with this geometrical foundation he applied to other
concepts, would place statistics upon a most firm foundation.
Fishers efforts received scant support in England, but he found a kin-
dred spirit in Hotelling. Even before Hotellings visit in 1929, the two
began to correspond. Their letters continued, as did Hotellings writings.
In 1930, in Transactions of the American Mathematical Society, Hotelling
presented The Consistency and Ultimate Distribution of Optimum Statistics;
Hotelling presented to a more receptive world, and with greater success,
concepts that Fisher had been struggling to produce.
17 Harold Hotellings Great Idea 169
There are a few scholars who are claimed by multiple disciplines. Fisher
and von Neumann were two striking examples. The same rare ability
to span multiple disciplines was demonstrated by Harold Hotelling. He
began as a theoretical mathematician who soon found application in eco-
nomics and in statistics. His location theory and his theory of exhaustible
resources are still taught in economics. And, he represented the establish-
ment of statistics as a discipline that now stands on its own, but spawned
from the unique combination of theoretical mathematics and practical
research methodological studies.
Over his career, Hotelling left an incredible legacy, both in statistics
and in the development of economics and finance. We know him for
his ubiquitous confidence intervals. Within the discipline of econom-
ics, he wrote the incredibly imaginative paper Stability in Competition1
in 1929 that initiated the theory of location analysis. His seminal 1931
work The Economics of Exhaustible Resources2 was the first rigorous work
in both resource economics and, with the work of the Great Mind Frank
Plumpton Ramsey (22 February 190319 January 1930), the introduc-
tion of decision-making over time, that is the defining innovation of
Notes
1. Hotelling, Harold, Stability in Competition (PDF), Economic
Journal 39 (153), 1929, pp.4157.
2. Hotelling, Harold, The economics of exhaustible resources. Journal
of Political Economy (The University of Chicago Press via JSTOR) 39
(2), April, 1931, pp.137175.
3. Hotelling, Harold, Edgeworths taxation paradox and the nature of
demand and supply functions. Journal of Political Economy, 40 (5),
October, 1932, pp.577616.
4. Hotelling, Harold, Demand functions with limited budgets.
Econometrica, 3 (1), January 1935, pp.6678.
5. Madow, W.G., Harold Hotelling as a Teacher, The American
Statistician 14 (3), 1960, pp.1517.
Part 4
The Birth of a Commission and
Econometrics
By the 1930s, statistics had been evolving over a little more than a century
and a quarter since Gauss least squares and Laplaces central limit theo-
rem. First as a tool for astronomical observations and global navigation,
it had been adopted by Galton and Pearson as a method by which bio-
logical characteristics could be measured and compared. This innovation
resulted in the creation of the journal Biometrika, with the express goal of
improving the measurement and analysis of scientific data, but especially
in biology. This evolution resulted in a revolution in statistical analysis,
but which was still confined to biological and agricultural questions.
Fischer had determined that modern statistics could not blossom based
merely on the clever characterization and aggregation of measurements
alone. Nor could it be built on the mere strength of personality of one
underfunded and academically marginalized, albeit brilliant, individual.
Perhaps more than any single person, Hotelling helped spark the statistics
movement on a more receptive side of the Atlantic Ocean. But, despite
these contributions in science and statistics, the tools did not treat the
problems most common in finance. It would take an entire team, some
deep pockets, and a few European scholars and war refugees to develop
the special tools needed for the peculiar challenges finance and econom-
ics data present. From these developments emerged the first society to do
for economics and finance what Pearson, Fisher and Biometrika did for
the sciences. It also resulted in a new journal, Econometrica. We next turn
to the story of those who made this possible.
19
The Early Life ofAlfred Cowles III
Some Great Minds make their mark through shear brilliance. Others
make a historically significant contribution not by solving a seemingly
intractable problem, but by recognizing, perhaps from a unique vantage
point, that a problem exists, or by offering, from their access to unique
resources, a pathway to solve the problem. Alfred Cowles III was a great
mind because of his unique ability to motivate others. He also brought to
the problem an infectious entrepreneurial spirit, and access to financial
resources almost never found among modern academics.
His contributions should come as no surprise.
The family name can be traced back to the fifteenth century. Sir John
Cole (14111500) haled from Nythway, Devon, England, where he was
married to Jane Merlot (14151500). They passed their peerage to Sir
William Cole (143989). Sir Williams son, John Cole (14701525),
survived his father and his mother, Elizabeth Weston (145589), and was
survived himself first by Archdeacon Thomas Cole (149410 May 1571),
his wife, Elizabeth Hargraves (14981574), and their son, William Cole
(152416 February 1600). By the birth of William, and his wife, Anne
Colles, the family had moved from Blade, Devon, where they spawned
John Cole
Samuel Cowles b: 1598 in London, London,
b: 1639 in Hartford,
Captain Isaac Cowles m: 14 Jun 1660 in
b: 28 Mar 1675 in d: 17 Apr 1690 in Hannah Hart
Farmington, Hartford, b: 1613 in Lilborne Parish,
Connecticut, United States
m: 27 Dec 1716 in
Farmington, Hartford,
Abigail Stanley
Ezekiel Cowles Connecticut, United States
d: 07 Feb 1756 in b: 17 Jun 1639 in Hartford,
b: 04 Nov 1721 in Farmington, Hartford, Hartford, Connecticut
Farmington, Hartford, Connecticut, United States d: 20 Oct 1734 in Farmington,
Connecticut, United States Hartford, Connecticut
m: 02 Sep 1752 in
Farmington, Hartford,
Elizabeth Smith
Connecticut, United States
d: 23 Sep 1806 in b: Feb 1690 in Farmington,
Farmington, Hartford, Hartford, Connecticut, United
Connecticut, United States States
d: 16 Aug 1767 in
Farmington, Hartford,
Giles Hooker Cowles Connecticut, United States
b: 26 Aug 1766 in
Farmington, Hartford,
Connecticut, United States
m: 05 Feb 1793 in of Bristol,
Conn. Giles Hooker
d: 05 Jul 1835 in Austinburg, b: 04 Oct 1690 in Farmington,
Ashtabula, Ohio, United Hartford, Connecticut, USA
States m:
d: 19 Feb 1787 in
Martha Hooker Farmington, Hartford,
Connecticut, USA
b: 03 Aug 1734 in
Farmington,Hartford,Connecti
cut,USA
d: 29 Nov 1817 in
Farmington, Hartford,
Martha Cooke
Connecticut, USA
b: 02 Jun 1693 in Hartford,
Hartford, Connecticut, USA
d: 22 May 1760 in
Farmington, Hartford,
Connecticut, USA
Shortly after their marriage, but before their western migration, Giles
and Sally had a son, Edwin Weed Cowles (3 May 17948 June 1861), the
first of their five boys and four girls. Born in Bristol, Connecticut, near
Farmington, his ancestors on both sides of the family were all Puritan
settlers, including a maternal ancestor Peregrine White, the first white
child born in on the Mayflower on its way to New England. Educated at
the Farmington Academy, Edwin Weed Cowles followed his family West
and became a prominent physician in Ohio. He and his wife Almira
Mills Foote (28 June 17889 April 1848) remained well established in
Ohio and raised children of great prominence.
Another son, Henry Cowles, followed his father into the ministry and
was one of the first professors at Oberlin College. A daughter, Betsey
Cowles (9 February 181025 July 1876), the eighth of ten children,
became an advocate for education reform and for the advancement of
women and abolition of the slave trade.
Edwin Cowles and Alfred Cowles, two other sons of Giles and
Sally, married two sisters from Cayuga County, New York, Elizabeth
Hutchinson (18271910) and Sarah Hutchinson (183784).
The first son of Edwin and Elizabeth Cowles, Edwin Cowles Jr.
(182590), had, at a young age, became an apprentice printer and co-
formed Smead & Cowles Printers. In 1853, he co-founded the Forest
City Democrat, and later changed its name to the Cleveland Leader news-
paper. He also helped found the Ohio Republican party out of the press-
room at the Leader. When he bought out his partner in the newspaper,
Mr. Medill relocated to Chicago and founded the Chicago Tribune with
Edwins brother Alfred.
Dr. Edwin Weed Cowles and his wife also had another son destined for
prominence. Alfred Hutchinson Cowles had studied geology, physics and
chemistry at the Ohio State University and Cornell University. He was
a student athlete who raced and beat Harvard in the eight-oared crew,
and raced at Henley, England. Upon graduation from his studies in geol-
ogy, he, his father and a brother founded the Pecos Mine in the Cowles
Mining District of New Mexico, which would become one of the richest
zinc, copper, silver and gold mines in the world.
There, Alfred Hutchinson Cowles and his brother Eugene Hutchinson
Cowles (185592) developed many new innovations in mining and
182 The Econometricians
Ednah Parker
b: 18 Oct 1776 in Bradford, Essex,
Massachusetts, United States
d: 07 Feb 1846 in Exeter,
Rockingham, New Hampshire,
United States
four young boys: Alfred (Bob) Cowles III, Knight Cheney Cowles (27
December 18921 May 1970), John Cheney Cowles (18941972) and
Thomas Hooker Cowles (6 June 189521 August 1927).
In 1916, just a few years after his graduation, and afflicted with tuber-
culosis, Bob too was attracted to the therapy that had been demonstrating
an improvement in its cure rate since it took the life of his mother. Under
the care of Dr. Gerald Webb, the renowned tuberculosis researcher, of
Colorado Springs, Bob recovered enough over a years stay at a sanitarium
to move in with his aunt Sarah Cowles Stewart and her husband Phillip
Stewart on Colorado Springs Millionaire Row.
20
The Times ofAlfred Cowles III
and resorts around the private Cheyenne Mountain Country Club and
the exclusive Broadmoor Hotel.
The young couple were fully immersed in the millionaire country club
scene. They partied and entertained, and they made a most handsome
couple. They also looked forward to starting a family together, but dis-
covered they could not have children. Instead, they traveled to Boston
to adopt, first, Richard Livingston, and then, a few years later, Ann. The
larger family also moved to a larger home, at 1506 Culebra, in Colorado
Springs.
At first, Cowles considered himself a journalist in the family tradition,
but he gradually became interested in financial investing, of which his
extended family had much to invest. He was the principal in an invest-
ment company, and from there, began his lifes work. Meanwhile, his
wife became increasingly close to Bobs aunt, and her husband Philip, the
son of a former Vermont governor.
Betsys diversion, first through her uncle by marriage, and then through
her father, became her emotional salvation. She was a young, capable,
well-liked and spontaneous woman, but she found herself married to a
man who was incredibly focused, to the point that he appeared to some
to be aloof and uninterested. Some speculate his focus arose from days
and nights on end in solitude and cold air as he fought for his life against
tuberculosis. His intensity was a great quality for one recovering from
a debilitating disease, for an investor and aspiring economist, and for a
financial theorist. It might not have been the ideal personality for a man
married to a vibrant woman much younger than he was.
Betsys salvation was her fathers newfound fascination with mountain
climbing. He had suffered physical problems later in life that led his doc-
tor to proclaim that he may lose the ability to walk. He took that diagno-
sis as a challenge to climb mountains. He brought his daughter into that
world, at the highest levels around the world. Betsy became a recognized
mountain climber, in Colorado and abroad.
In the 1930s, Betsy fell out of love, and the couple divorced in 193839.
That year, Cowles moved back to Chicago, but continued to support Betsy
for the rest of her life, and continued to manage her fathers investments.
Ten years later, on 24 October 1949, Cowles remarried, to Louise Lamb
Phelps. Betsy later married an air force general, Earle (Pat) Partridge.
20 The Times ofAlfred Cowles III 189
With his interest in statistics stimulated and the time afforded with a life
of semi-retirement, Cowles became interested in applying statistical prin-
ciples to a finance discipline for which he had developed a professional
suspicion that it lacked scientific rigor. He had become growingly suspi-
cious of the spurious advice offered by many financial advisors over the
Roaring Twenties, especially in the wake of the Great Crash of 1929, and
he had schooled himself on the mathematics of least squares and statistics
that could act as the basis for the treatment of finance as a science rather
than an art. His first project in finance was to determine the effective-
ness and correlation of the advice from 24 market newsletters and stock
market performance.
Of course, Cowles could have afforded to commission economists and
statisticians to do his analytic work for him, but he also had the time and
interest to take a hands-on approach. To assist him in the statistical theory,
he came upon an academic, Harold Thayer Davis (5 October 189214
November 1974), who spent summers in the mountain community of
Colorado Springs, Colorado. Davis knew that while the regression calcu-
lations could be performed by a team of clerks, he thought a new device
emphasis from the cold world of neo-classical economics that was the
rage at the time to a new world of forecasting and better public policy.
By the early 1930s, Cowles had assembled much of the mathematical,
statistical and computing resources he would need. He then pressed on
his Yale connections to woo the economics discipline. At that time, the
most prominent American economist was the Great Mind Irving Fisher
(27 February 186729 April 1947). A Yale graduate and faculty member,
he had known Cowles father from college days. In fact, both Cowles
father and uncle were Yale graduates, as were others who followed him.
Fisher was highly prominent, if not partially disgraced by his very public
optimism, of the strength of US financial markets even right up to and
immediately following the Great Crash.
Fisher himself had built up a financial fortune from his invention of
data sorting techniques, one of which was an early version of the once
ubiquitous Rolodex. While Fisher irretrievably lost his fortune with the
onset of the Great Depression, he was nonetheless a powerful figure who,
by chance, shared another important characteristic with Cowles. Fisher,
too, had suffered tuberculosis as a young man, and had even spent some
time in Colorado Springs. These connections afforded Cowles a receptive
ear from one of the most well-known and respected economists in the
country who had, for years, been advocating the formation of a society
devoted to scholarly work in econometrics.
Cowles offered the funding if Fisher could assemble the critical
mass of scholars both to create a research commission and to form the
Econometric Society that would rival the statistical societies of England.
Fisher brought the academics, and Cowles built the facility, first in
Colorado Springs.
One motivation of Cowles was his desire to produce a statistical
study of whether stock market forecasters are successful in predicting
the direction of the market. He believed they did not. Cowles devoted
his profound intensity into the analysis of almost five years of data from
7500 investment advisor recommendations from 16 services. From his
mammoth study, he demonstrated that only 6 of 16 services provided
positive returns, with the average investment service actually yielding
financial losses.
194 The Econometricians
His analysis was read at one of the first meetings of his new Econometric
Society, and was published in one of the first editions of the new jour-
nal Econometrica he funded. His Cowles Commission for Research in
Economics, with the motto Science is Measurement, quickly became
the nations most prominent financial and economic research group.
Cowles proved to be one of the Commissions most insightful research-
ers early on. His interest began with his stock market research. Like
Fisher, who had been for decades collecting data on the inflation rate,
bond prices and bond interest rates, with the goal of developing finan-
cial instruments that are immunized from inflation and money supply
manipulation, Cowles realized that data is king. He set about produc-
ing stock price databases that would eventually become the Standard &
Poors 500 index. His indices were much more sophisticated and theo-
retically sound than the simplistic Dow Jones Industrial Index that was
commonly employed at the time.
From his database of stock prices and forecasts, Cowles created a set
of 24 time series of forecasts and demonstrated, within the context of
a multivariate regression, that most of these forecasts were not statisti-
cally correlated to market success. In other words, most forecasts perform
no better than random choices. He stated, The most successful records
are little, if any, better than what might be expected to result from pure
chance.1
In effect, Cowles was developing an early version of the random
walk that was popularized by the Great Mind Paul Samuelson (15 May
191513 December 2009) more than 30 years later, in 1964. It was this
first volley into the premise that forecasters cant beat the market that
produced what now underpins much of financial theory, and was the
subject of an earlier volume in this series on the efficient market hypothesis.
While Cowles emerged as a competent self-taught scholar, his con-
tribution hardly stopped there. Cowles was an academic benefactor and
entrepreneur perhaps without parallel. Never a professor himself, and
having never enjoyed the trappings of the Ivory Tower, Cowles none-
theless successfully assembled one of the greatest groups of scholars in
academic history.
Aside from some early contributors from the mathematical and comput-
ing sciences, Cowles very quickly identified a number of individuals who
he felt could help further the creation of a fledging econometric society.
21 The Great Idea ofAlfred Cowles III 195
the USA.Cowles and his Commission found itself on the leading edge
of both and with the resources and stature to attract the worlds leading
scholars.
Another aligning of the stars at the time was the acceleration of
nationalism under Adolf Hitler in Germany, and his warped application
of Galtons eugenics. Some of the brightest applied mathematicians in
Germany, Austria and Eastern Europe at the time were of Jewish descent.
Many were counseled by colleagues to not risk their fate to an increas-
ingly militant German movement, and some others began to feel less
subtle persuasions to leave Europe.
One of the earliest and most formative members of the Cowles
Commission was one such European scholar, Jacob Marschak (23 July
189827 July 1977). He, Ragnar Frisch and Haavelmo, all joined the
Cowles Commission, which, by then, had moved to the University of
Chicago to be better able to tap the resources of one of the countrys best
economics departments, and, by no small coincidence, the family home
of Cowles.
Marschak was a most colorful character. When he joined the Cowles
Commission as its research director in 1943, he joined scholars whose
families had followed a similar migration from regions in and around
the Austro-Hungarian Empire, and found themselves at Cowles, includ-
ing the families of Milton Friedman (31 July 191216 November 2006)
and Franco Modigliani (18 June 191825 September 2003). Marschak
found himself as an immigrant who sought academic refuge at the New
School for Social Research in NewYork City after a four-year stint direct-
ing Oxfords statistics institute in the late 1930s. He had joined the same
University in Exile that Harold Hotelling had supported. Many of these
almost 200 scholars were brought there by the New School co-founder,
Dr. Alvin Johnson (18 December 18747 June 1971), between 1933 and
1945.
First at the New School, and then at Cowles, Marschak found himself
surrounded by the same eclectic and intellectually diverse scholars who
could look at old academic problems in new ways and with new tech-
niques. If they did not have the new techniques yet, they created them.
Marschak and his colleagues at Cowles recognized that the static model
which predated the Great Depression was ill-prepared to treat the rapidly
21 The Great Idea ofAlfred Cowles III 197
Note
1. Cowles, Alfred, III, Can Stock Market Forecasters Forecast?
Econometrica 1(3), July 1933, pp.309324.
22
Legacy andLater Life ofAlfred
Cowles III
While Alfred Cowles III was not college-trained in finance and economics,
he nonetheless attained some of the highest accolades of his field. He was
a Fellow of the American Association for Advancement of Science, and a
Fellow and Treasurer of the Econometric Society. He also was the princi-
pal author of the Cowles Monograph Common Stock Indexes, and contrib-
uted to Econometrica and the Journal of American Statistical Association.
Some of his publications include Can Stock Market Forecasters Forecast?,
a paper read before a joint meeting of the Econometric Society and the
American Statistical Association, Cincinnati, Ohio, 31 December 1932.
This article was subsequently reprinted in Econometrica.1
Cowles also wrote Analysis of 4 Years of Forecasting by 41 Advisory
Services and Publications,2 and wrote, with Herbert E. Jones, Some a
Posteriori Probabilities in Stock Market Actions.3 Other Econometrica papers
included Stock Market Forecasting,4 and A Revision of Previous Conclusions
Regarding Stock Price Behavior.5
His Cowles Commission Monographs included Common-Stock
Indexes, 18711937,6 Common-Stock Index, 2nd edition,7 and Supplement
to Common-Stock Indexes for 1939 and 1940.8
Notes
1. Cowles, Alfred, III, Can Stock Market Forecasters Forecast?
Econometrica 1(3), July 1933, pp.309324.
2. Cowles, Alfred III, Analysis of 4 Years of Forecasting by 41 Advisory
Services and Publications, Stock Market Technique, ed. by Richard
D.Wychoff, Vol. 2, No. 2, April 1933.
3. Cowles, Alfred III and Herbert Jones, Some a Posteriori Probabilities
in Stock Market Actions, Econometrica, (July 1937), 5(3): 280294.
4. Cowles, Alfred III, Stock Market Forecasting, Econometrica, July
October 1944, pp.206214.
5. Cowles, Alfred III, A Revision of Previous Conclusions Regarding
Stock Price Behavior, Econometrica (October 1960), 28(4):
pp.909915.
6. Cowles, Alfred III, Common-Stock Indexes, 18711937, 1st edition.
Bloomington: Principia Press, Inc. Bloomington, 1938.
7. Cowles, Alfred III, Common-Stock Index, 2nd edition, Principia Press,
Inc., Bloomington, 1939.
8. Cowles, Alfred III, Supplement to Common-Stock Indexes for 1939 and
1940, University of Chicago Press, Chicago, 1940.
23
The Early Life ofRagnar Frisch
While Cowles and his scholarship were distinctly American, two of the
greatest influences on the Cowles Commission were distinctly Norwegian.
This tradition began with Ragnar Frisch.
The family name Frisch is German in origin, but the family has lived
in Norway for almost four centuries.
In the early summer of 1623, two children were herding cattle along a
pathway on Gruvesen hill near present-day Kongsberg, Norway. At that
time, the town, located about 50 miles west and south of Oslo, was part
of the Kingdom of Norway and Denmark, under the monarch Christian
IV.The ox the children were using to help herd their cattle rubbed against
the mountainside and revealed a large shiny stone. The children, Helga
and Jacob, brought the shiny stone to their father, who melted it down
and took it into the nearby town of Skien in Telemark County to sell. He
fetched such a low price for it that the local police became suspicious it
might be stolen. The father was given a choiceto be arrested to tell the
authorities where he found what turned out to be pure silver from what
would become one of the worlds richest mines.
Ragnar Frisch
b: 03 Mar 1895 in Oslo,
Norway
m:
Herr Kittilsen
d: 31 Jan 1973 in Oslo,
Norway b:
m:
d:
Ragna Fredrikke
Kittilsen
b: 1860 in Norway
d: 1936 in Oslo,
Norway
By the time Frisch began to lead the development of the new field of
econometrics, Ronald Fishers Handbook had been well reviewed by Harold
Hotelling, and statistics groups were beginning to sprout across the USA,
especially at the major land-grant universities that contained in their mission
the need to perform research in support of agriculture. With a fresh PhD in
hand by 1926, Frisch was about to join that fray among other founders of
econometrics. In 1927 he was offered a Rockefeller Foundation scholarship
to travel to the USA.One of his first collaborations came when he arrived at
Yale University in New Haven, Connecticut, to work with the great mind
Irving Fisher, just as Cowles was encouraging Fisher to bring to fruition the
vision of an Econometric Society.
This year in the USA proved most influential for both Frisch and sta-
tistics and finance. As discussed earlier, it was his meeting with Fisher that
gave rise to the new Econometric Society that Alfred Cowles III would
later underwrite.
Many of the leading finance, economics and statistical theorists of
the day were trained first as physicists or engineers. This included the
great mind Irving Fisher, and many of his statistics contemporaries,
from Ronald Fisher to Harold Hotelling. Frisch shared that belief the
Roos agreed that there was a need to develop an Econometric Society, and
decided to travel to Fishers home in April of 1928 to solicit his help.
Fisher was at first pessimistic about the possibilities as his efforts to do
the same over the past two decades had garnered little momentum. Fisher
asked Frisch and Roos to name a hundred scholars who would be willing
to participate. Perhaps he surmised that Frisch, from Europe, and Roos,
from mathematics, would be hard-pressed to reach such a critical mass
for a new Society.
Frisch and Roos came up with a number of names very quickly, but,
after only a few days, hit upon the wall economists label diminishing
marginal returns. When they came up with only about 70 names, Fisher
quickly added a dozen of his own, and the three men agreed that they had
enough to begin their joint efforts.
In the winter of 1930, the three main associations, the American
Economic Association, Section K of the American Association for the
Advancement of Science and the American Statistical Association all met
in Cleveland. Earlier in 1930, Frisch, Roos and Fisher had sent out invita-
tions to attend a meeting while in Cleveland. Frisch, who had returned to
teach at the University of Oslo following the completion of his Rockefeller
Foundation scholarship, happened to be in the USA as a visiting profes-
sor at Yale, was in attendance in Cleveland, and joined Roos, Fisher and
13 other scholars from the USA and Europe at the Cleveland hotel on 29
December 1930 to inaugurate the new Econometric Society.
Joseph Alois Schumpeter (8 February 18838 January 1950), then of
the University of Bonn, chaired the meeting, and Fisher was elected its
first president. Frisch produced the draft of the Societys constitution,
and a series of meetings, in Lausanne, Switzerland, in September of 1931,
and Washington, DC, in December, were agreed upon. By the end of
that year, 173 members were elected.
Cowles approached the new society on the behest of his Colorado
Springs colleague, Harold T.Davis, the mathematics professor at Indiana
University who spent summers in Colorado. Davis knew of the fledging
society and thought their interests aligned perfectly with Cowles desire
to bring measurement and science to economics and finance. Fisher was
elated to see that Cowles could help resolve the Econometric Societys
lofty goals but meager dues and budget.
208 The Econometricians
to educate and benefit its members and mankind, and to advance the sci-
entific study and development of economic theory in its relation to
mathematics and statistics.
Cowles was elected a trustee, a laboratory was renovated and the eco-
nomics faculty of Colorado College were enlisted to create the organiza-
tion necessary to pursue its first projects. Cowles provided an annual
budget of $12,000.
The inaugural issue of the journal Econometrics came out in the first
month of 1933, with Frisch framing its editorial. Schumpeter offered the
following introduction to the journal:
The common sense of the program of our Society centers in the question:
Can we not do better than this? Surely it would not be a reasonable policy
to sit down and wait till, in the end, things find their level by themselves,
and meanwhile to allow econometricians of all countries to fight single-
handed their uphill battle. What we want to create is, first, a forum for
24 The Times ofRagnar Frisch 209
econometric endeavor of all kinds wide enough to give ample scope to all
possible views about our problems, yet not wide enough to be hampered by
the weight of an audience which keeps discussion in the ante-rooms of the
real points at issue, and forces every speaker or writer to go every time over
the same preliminaries.
that it leads right up to the core of the burning questions of the day, but
rather because we expect, from constant endeavor to cope with the difficul-
ties of numerical work, a wholesome discipline, the suggestion of new
points of view, and helps in building up the economic theory of the future.
But we believe, of course, that indirectly the quantitative approach will be
of great practical consequence. The only way to a position in which our
science might give positive advice on a large scale to politicians and busi-
ness men, leads through quantitative work. For as long as we are unable to
put our arguments into figures, the voice of our science, although occasion-
ally it may help to dispel gross errors, will never be heard by practical men.
They are, by instinct, econometricians all of them, in their distrust of any-
thing not amenable to exact proof.2
Notes
1. Frisch, Ragnar, Sur un problme d'conomie pure (On a problem in pure
economics). Norsk Matematisk Forenings Skrifter, Vol. 1, No. 16, 140, 1926.
2. Schumpeter, Joseph, The Common Sense of Econometrics, Econometrics,
1, 1, 1932, pp.512.
25
Ragnar Frischs Great Idea
The rousing editorial of Joseph Schumpeter was not the only immortal
contribution to the first edition of Econometrica. Also in the volume
was a paper by Ragnar Frisch and his agricultural economist colleague
F.W.Waugh that both postulated and solved a problem that is both some-
what unique and ubiquitous in financial and economic data.
To then, Frisch was well versed in the methods of mathematical sta-
tistics in his day, but also appreciated the challenges of extending these
same tools to financial analyses. Frisch especially tried to understand the
vagaries of the business cycle. It is in this context that he differentiated
between the microeconomics of individual markets, for which much of
the pre-existing economics literature was devoted, to the macroeconom-
ics of the aggregate economy, which often suffers in opaqueness due to
the very aggregation. A few years later, the Great Mind John Maynard
Keynes (5 June 188321 April 1946) offered an alternative model of the
macroeconomy and, for the first time, successfully explained the poten-
tial persistence of economic depressions.
In fact, some of Keynes concepts on the functioning of the macro-
economy seem to follow from Frisch. By 1929, just as global econo-
mies were just showing hints of vulnerability, Frisch was one of the first
If, on the other hand, he designs the analysis so as to make the time
sequence of the cards a factor of special importance, the analysis is dynamic.
In the former case, therefore, time is just a kind of auxiliary variable. In the
latter case, the time pattern itself is of central importance.1
depressions. From his analysis, he was able to demonstrate, for the first
time, that random shocks to the economy could produce wave-like busi-
ness cycle fluctuations.
From these experiences, Frisch was one of the first among the Cowles
Commission scholars to recognize that the tools of mathematical sta-
tistics as developed for random observational errors in astronomy and
biology must be adapted for the special challenges of econometrics and
finance. The statistics to then assumed that errors were drawn from an
identical, often normal, distribution, and each was independent of other
measurement errors. In other words, there were no discernable trends in
the errors.
Second, it is assumed that the explanatory variables were orthogonal
to each other. In other words, these variables were uncorrelated with each
other. However, in time series data, it is often the case that the size of
errors grows as time progresses and the size of the explanatory variables
increases, which causes a problem called heteroscedasticity. Also, a time
trend shared by the various explanatory variables may cause them to be
correlated with each other.
In the latter case, it is a challenge to attach this common trend to any
one of the explanatory variables. Frisch asked if this problem would then
distort the coefficients and their reliability associated with the estimations.
In a brilliant paper published with the American agricultural econo-
mist F.V.Waugh (18981974) in the very first edition of Econometrica
in 1933, Frisch resolved this dilemma. He demonstrated that the coef-
ficients of estimation in a linear least squares regression remain the same
if not detrended of time, or if detrended separately. This proof lent confi-
dence to scholars using the primary tool of the day to analyze the prevail-
ing economic and finance problems of the day.
Consider a linear regression model in which a set of dependent obser-
vations Yt over time are postulated to depend on a set of independent
variables Xt over time, where the independent variable vector at each
point in time also includes the constant unity term. Let there also be a
time trend t. Then, the relationship is postulated to be:
Yt = b Xt + a Tt + t
214 The Econometricians
Frisch and Waugh proposed that the time trend be first removed sepa-
rately from both the dependent and the independent variables through a
series of simple linear regressions:
Xit = ci 0 + ci1T + itx
Yt = c0 + c1t + . y
t
Xi*t = X i + e ixt
Yt* = Y i + e ty .
Yt* = b * Xt* + e t* .
Frisch and Waugh showed, and Lovell extended the proof, that the
coefficients and * will be identical in either case. This provided the
justification necessary to apply the various tools of mathematical sta-
tistical regression modeling to the more complicated realm of finance
and econometrics, and allowed a vast expansion of the foundations of
econometrics.
To see how Frisch and Waugh proved this, and how Lovell extended
the method in 1963 to also include the use of dummy variables, lets
consider a simple ordinary least squares regression with only one inde-
pendent variable. Let,
25 Ragnar Frischs Great Idea 215
Yt = b X t + a Tt + e t .
We will compare the coefficient from this regression with the corre-
sponding coefficient * from the following regression without the explicit
common trend:
Yt = b * X t + e t* .
Y = aY T + e Y
X = aXT + e X .
Y = b X + aT + e
Y = aY T + e Y
X = aXT + e X .
b X + a T + e = aY T + e Y .
216 The Econometricians
Substituting into the equation above, the expression for X then gives:
b (a X T + e X ) + a T + e = a Y T + e Y .
This first property implies that both X and Y are uncorrelated with
the trend variable T. Then, the second property implies that all the coef-
ficients in the brackets of the collected terms equal zero, and:
e Y = be X + e .
we find that:
Y + eY = b X+ e X + e
Y = b X + e.
* *
25 Ragnar Frischs Great Idea 217
Notes
1. Frisch, Ragnar, Statikk Og Dynamikk I Den konomiske Teori,
Nationalkonomisk Tidsskrift, Bind 3. rkke, 37, 1929, translated as
Statics and dynamics in economic theory, Structural Change and Economic
Dynamics, 3, 1992, pp.391401, at 323.
2. Frisch, Ragnar, Propagation Problems and Impulse Problems in Dynamic
Economics. (1933) In Economic Essays in Honor of Gustav Cassel. Reprinted
in R.A. Gordon and L.R. Klein, eds., Readings in Business Cycles. London:
Allen & Unwin, 1966.
3. Lovell, M., Seasonal Adjustment of Economic Time Series and Multiple
Regression Analysis. Journal of the American Statistical Association 58 (304),
1963, pp.9931010.
26
Legacy andLater Life ofRagnar Frisch
With this lofty beginning, and the combination of Frischs vision and
tenacity and Cowles funding, came to be both a new scholarly society
and perhaps the greatest research commission ever created in the social
sciences.
But, despite the great distance between Colorado Springs and Oslo,
Norway, Frisch nonetheless remained engaged in the journal and influ-
ential in the Commission. His American kindred spirit Roos was the
Commissions first director. Roos moved on in 1937, and was temporarily
replaced by Cowles Colorado Springs colleague Harold Davis. Davis and
Roos each helped direct Cowles and the Commission for the next couple
of years. Then, in 1938, the Commission convened Frisch, Marschak and
the University of Chicago business professor Theodore O.Yntema to try
to entice any of them into directing the Commission. Each felt Colorado
Springs isolation would be academically problematic.
Then, with the death of Cowles father in January of 1939, Alfred
Cowles III was required to relocate back to his familys business base in
Chicago. On 29 September 1939, one month short of the tenth anniver-
sary of the Great Crash that motivated much of the ensuing discussion,
Frisch suffered first the vagaries of the Great Depression, and then
World War II in ways few can imagine. Indeed, he was fortunate to sim-
ply survive the war, though. In October of 1943, the Nazis, who had
occupied Norway, interred Frisch to a concentration camp, where he
would spend a year. He shared this insult with a number of other profes-
sors and associates of his Institute.
Upon his release from the Nazi prison in the autumn of 1944, he was
still unable to communicate with the USA through regular channels for
another year. One of the first correspondences Frisch received was in July
of 1945. A letter from his prodigal student, Trygve Haavelmo, described
success Haavelmo was having in his residency in the USA during the
war years. Haavelmo had just published a major work on the probability
approach to econometrics, and noted:
I must have been quite likely with (the article) as the Cowles Commission
with Marschak in charge, took up the methods suggested in the article and
now has a special group working on further corroboration I attended a
conference about this work in Chicago this winter. Hotelling and others of
the big boys were also invited, and it was thus quite fun.
Frisch was a devoted and generous mentor, and was pleased to see
his young prodigy escape the fate of imprisonment, and, at the same
time, help redefine modern econometrics. He recognized that neither
economics nor politics can solve social problems in isolation. Instead, the
political bodies must work with the learned societies to solve problems.
This was consistent with his Scandinavian colleague Knut Wicksell (20
December 18513 May 1926), who established the Stockholm School
in this mold. Frisch devoted much of his life after the war to assist the
Norwegian government in these lofty technocratic goals.
His devotion to a mix of practical and theoretical problems earned Frisch
a number of recognitions. He received the Accademia Nazionale dei Lincei
award in 1961, and the first Nobel Memorial Prize ever awarded, in 1969,
which he shared with Jan Tinbergen (21 April 19039 June 1994). He also
left perhaps more unfinished manuscripts than ones he published. He was a
notorious perfectionist, which meant he often failed to polish and publish
work that lesser scholars would have nonetheless submitted.
222 The Econometricians
Frisch edited the journal Econometrica for 21 years. The annual Frisch
Award is offered every two years in his honor for the best paper in
Econometrica. There is also a Frisch Medal, and a Frisch Center named in
his honor at the University of Oslo.
He enjoyed a fulfilling personal life as well. In 1920, he had married
Marie Smedal, and they shared a child, Ragna. His wife died in 1952,
and he subsequently married an old college friend, Astrid Johannessen.
She had graduated from the University of Oslo his same year. The daugh-
ter of I.M.Johannessen, a shipping magnate from masted ship days, and
his wife Julie Caspersen, Astrid grew up in relative comfort. These friends
of Ragnars parents had produced a daughter for whom Ragnar was lov-
ingly devoted.
Together, Ragnar and Astrid enjoyed Ragnars granddaughter, his
interests in mountain climbing and beekeeping, and his amateur stud-
ies of the eugenics of bees. He admitted his interest in genetics, and its
related statistics may have been more of an obsession than an avocation.
27
The Early Years ofTrygve Haavelmo
(191384) and Trygve (13 December 191128 July 1999) retained the
same Haavelmo family name (Fig. 27.1).
Their father, Halvor, grew up on a farm named Hvlmoen in Gol,
but had the opportunity to complete school and attend the Elverum
Teacher Training College. He passed the teaching exams in 1907 and
took a teaching job in his village of Gol. A year later, he took a job in
Skedsmo, about ten miles northeast of Oslo, where he spent the rest
of his career.
Halvor was a successful and well-regarded teacher and principal in
Skedsmo. He also was appointed mayor of Skedsmo for two terms,
and sat for many years on its school board. He maintained an interest
in athletics and encouraged sports for his sons as well. The younger
son, Sverre, was a champion ski jumper at a ski hill Halvor helped
establish.
Halvor became the town historian and published a number of volumes
on local history. His children were provided every opportunity for a good
education.
27 The Early Years ofTrygve Haavelmo 225
In these years at Oslo, Frisch traveled and visited the USA extensively.
By 1936, Haavelmo was ready to pursue advanced study in statistics,
and attended University College London, where he studied and worked
directly with Egon Pearson, the son of Karl Pearson. During his stay at
the University College London, he was undoubtedly drawn somewhat
into the battle of the titans, Egon Pearson and Ronald Fisher. While
Haavelmo remained loyal to his mentor Pearson, and Pearsons brilliant
colleague, Pearsons co-developer of the likelihood ratio test, Haavelmo
maintained a strong geometric intuition about statistics, which was much
more closely aligned with Fisher.
Haavelmo also studied in Oxford, Berlin and Geneva, Switzerland,
before he took on his first academic position at the University of
Aarhus in Denmark. He remained there for a year before he was
afforded the opportunity to visit and network in the USA in 1939 on
a Rockefeller Foundation scholarship. Following completion of his
Rockefeller Foundation scholarship, he became associated with the
Cowles Commission, and remained in the USA to work on behalf of
the Norwegian governments Nortraship (Norwegian Shipping and Trade
Mission) in NewYork City for the remainder of the war.
28
The Times ofTrygve Haavelmo
Note
1. Christ, Carl F. (1994). The Cowles Commission Contributions to
Econometrics at Chicago: 19391955. Journal of Economic Literature 32
(1), 1994, pp.3059, at p.40.
29
Haavelmos Great Idea
assumptions. Indeed, recall that Gauss had derived the least squares
technique from a maximum likelihood perspective.
However, if one postulates that consumption is a linear function of
an intercept, likened to Keynes autonomous consumption term, and
an induced term that rises proportionally with income, and if, in turn,
income depends on the level of consumption spending, the simple linear
least squares model is violated because random errors arise both in the
simple equation and in the determination of income. Ronald Fisher had
treated such a generalization in his maximum likelihood statistics at the
beginning of the twentieth century. Haavelmo expanded this methodol-
ogy of maximum likelihood estimation to both highlight when the least
squares model creates biased estimations, and how the method of maxi-
mum likelihood can instead be employed.
As a brief example of Haavelmos insights, consider the interaction
between the aggregate supply of goods and services in an economy and
its aggregate demand. The supply comes from the aggregated decisions
of millions of market participants who formulate their decisions based in
part on the income they would earn from their sales, and the factor pay-
ments they must pay to the resources they would employ.
Another equation might describe how the recipients of these factor
payments use their income to purchase these same goods and services.
In equilibrium, these two equations of aggregate demand and aggregate
supply would intersect at an aggregate price and income.
The estimation of this system would provide perhaps a corresponding
income at the intersection of these two equations. But, one discerned
intersection point could actually be the product of many aggregate
demand and supply curves. The mere identification of the equilibrium
intersection point, which is subject to some random error, sheds little
light on the precise nature of the two underlying curves. Hence, a wide
variety of parameters for candidates for the underlying curves could be
consistent with the observed market variables.
This observation led to a new line of research that would become the
basis for a most fruitful period of research by Cowles Commission fel-
lows, and would lead these fellows to numerous Nobel Prizes. Among
them are Nobel Laureates that include Tjalling Koopmans, Kenneth
Arrow, Grard Debreu, James Tobin (5 March 191811 March 2002),
29 Haavelmos Great Idea 233
Where statistical data are used as one of the foundation stones on which
the equation system is erected, the modern methods of statistical inference
are an indispensable instrument. However, without economic theory as
another foundation stone, it is impossible to make such statistical inference
apply directly to the equations of economic behavior which are most rele-
vant to analysis and to policy discussion. Statistical inference unsupported
by economic theory applies to whatever statistical regularities and stable
relationships can be discerned in the data. Such purely empirical relation-
ships when discernible are likely to be due to the presence and persistence
of the underlying structural relationships, and (if so) could be deduced
from a knowledge of the latter. However, the direction of this deduction
cannot be reversed from the empirical to the structural relationships
except possibly with the help of a theory which specifies the form of the
structural relationships, the variables which enter into each, and any fur-
ther details supported by prior observation or deduction therefrom. The
more detailed these specifications are made in the model, the greater scope
is thereby given to statistical inference from the data to the structural equa-
tions. We propose to study the limits to which statistical inference, from
the data to the structural equations (other than definitions), is subject, and
the manner in which these limits depend on the support received from
economic theory.4
234 The Econometricians
During the 1930s, noteworthy attempts were made to test economic theo-
ries empirically. The results of these attempts called attention to two funda-
mental problems associated with the possibility of testing economic
theories. First, economic relations often refer to large aggregates of indi-
viduals or firms. Theories regarding such relations can never be expected to
conform fully with available data, even in the absence of measurement
errors. The difficult question then is to determine what should be consid-
ered sufficiently good, or better conformity. Second, economists can sel-
dom or never carry out controlled experiments in the same way as natural
scientists. Available observations of market outcomes, etc., are results of a
multitude of different behavior and relations which have mutually interact-
ing effects. This gives rise to interdependence problems, i.e., difficulties in
using observed data to identify, estimate, and test the underlying relations
in an unambiguous way.
29 Haavelmos Great Idea 235
The aim of econometrics, as you may read it on the cover of every issue of
Econometrica, is The Advancement of economic theory in its relation to
Statistics and Mathematics. That is, econometrics should be an attempt,
not only towards more precision in the formulation of economic theories,
but perhaps still more an attempt to reach such formulations that the theo-
ries lend themselves to testing against actual observations.
So far, however, it seems that most of the energy has been spent on con-
structing rational models, involving exact relationships that are much too
rigid if we would try to identify the theoretical variables involved with
some actually observable economic quantities.
The relation between such exact economic models and economic reality is,
of course, in point of principle similar to the relation between rational
mechanics and the bodies and motions observed in the real physical world.
But the difference in the degree of precision is and is probably always going
to be tremendous. In economics it is therefore not sufficient first to set up
a system of exact relationships and then allow for certain small deviations
in the applications of facts. We shall have to start out with a probabilistic
formulation of our models, from the beginning, otherwise we shall either
have to call all our theories wrong or we may call almost any theory right
by allowing for sufficiently large discrepancies (in a subjective manner).7
Haavelmo elaborated:
Notes
1. Frisch, Ragnar, Statistical confluence analysis by means of complete regression
systems, Universitetets , konomiske institutt, 1934.
2. Haavelmo, Trygve, The Statistical Implications of a System of Simultaneous
Equations, Econometrica, Vol. 11, Issue 1, 1943, pp.112. Bjerkholt, Olav,
Tracing Haavelmos steps from Confluence Analysis to the Probability
Approach, Memorandum 25/2001, University of Oslo Department of
Economics and the Frisch Center for Economic Research, May 2001, at p.29.
3. Koopmans, Tjalling C, Identification problems in economic model con-
struction. Econometrica 17 (2), 1949, pp.125144.
29 Haavelmos Great Idea 239
4. Ibid. at 126.
5. The Prize in Economics 1989 Press Release. Nobelprize.org. Nobel Media
AB 2014. Web. 16 March 2016. <http://www.nobelprize.org/nobel_prizes/
economic-sciences/laureates/1989/press.html>
6. Haavelmo, Trygve, The Probability Approach in Econometrics. Supplement
to Econometrica, Vol. 12, July, 1944, pp. S1S115.
7. Bjerkholt, Olav, Tracing Haavelmos Steps from Confluence Analysis to the
Probability Approach, Memorandum 25/2001, University of Oslo
Department of Economics and the Frisch Center for Economic Research,
May 2001, at pp.2728.
8. Ibid, at p.5.
9. Ibid. at p.3.
30
Legacy andLater Life ofTrygve
Haavelmo
Notes
1. The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred
Nobel 1989, Nobelprize.org. Nobel Media AB 2014. Web. 25 March 2016.
< http://www.nobelprize.org/nobel_prizes/economic- sciences/
laureates/1989/>
2. http://articles.latimes.com/1989-10-12/news/mn-294_1_economics-nobel-
prize, accessed 16 March 2016.
3. Bjerkholt, Olav, Tracing Haavelmos steps from Confluence Analysis to the
Probability Approach, Memorandum 25/2001, University of Oslo
Department of Economics and the Frisch Center for Economic Research,
May 2001, at p.30.
Part 5
What We Have Learned
This book is the seventh in a series of discussions about the great minds in
the history and theory of finance. It is somewhat unique in this series in
that the other treatments demonstrated a relatively straight line of inno-
vation, with each contribution by a great mind shoring up and expanding
the contributions of those who came before.
The Great Minds that gave rise to econometrics, though, are much
more varied. They included one of humankinds greatest mathematicians,
a charismatic explorer, an early biometrician and an agricultural statisti-
cian. From their work in an evolving statistics, an applied mathematician
in the USA, a financial entrepreneur and a pair of scholars from Norway
turn statistics into the econometric tools now used by financial theorists.
We conclude with a brief summary of their collective contributions.
31
Conclusions
the intuition of the central limit theorem, a result that his colleague and
friend Pierre-Simon, the Marquis de Laplace generalized in a way that
have provided us much intuition ever since. One cannot compete a foun-
dational tract in statistics within our university finance curricula without
an indoctrination of Gauss contributions.
Across the English Channel in the mid-nineteenth century, as Gauss
was reaching the eve of his life, an explorer and man of leisure with a
remarkable ability to sense the pulse of populism, Sir Francis Galton,
was creating in his own mind a theory of eugenics just as his famous
cousin Charles Darwin was revolutionizing the theory of evolution.
Galton needed some rudimentary statistical techniques to support his
hypotheses, and applied some of the insight of Gauss and his method
of least squares to demonstrate such tendencies that the height of sub-
sequent generations tends to regress toward the mean of their parents
heights. Galton was no mathematician, though, so it is entirely possible
that the statistical movement he fueled with his intuition was not initially
informed by Gauss brilliance that came before him.
It took Galtons successors, first Karl Pearson, and then Ronald Fisher,
to create first the measures of basic statistics, some of which Gauss had
developed a century later, and then the geometric interpretation for mod-
ern hypothesis testing, again by applying the techniques so masterfully
developed by Gauss.
The story told described a clash of personalities that created a toxic
statistical environment on one side of the Atlantic to the point that the
statistics discipline could not progress efficiently until it was effectively
transplanted to the other side of the Atlantic Ocean. In the USA in the
1920s and 1930s, Harold Hotelling advocated for the statistics and
the department that would develop them, and began to use the tool to
answer questions of economic and finance significance. But, Hotelling
had an active mind that drew him in many directions.
It would take a financial entrepreneur and practitioner willing to
devote a portion of his familys fortune to further develop and popularize
techniques that could be applied to modern finance. Alfred Cowles III,
called Bob by his family and friends, created a Cowles Commission for
which almost every leading econometrician from Ragnar Frisch to Harry
Markowitz and beyond was associated.
31 Conclusions 249
Cowles vision for his commission gave rise to the Econometric Society
and the journal Econometrics that would do for financial statistics what
Biometrika did for biological statistics. He also originated important con-
cepts himself, such as the notion of the random walk and the efficient
market hypothesis.
Our story ends with two individuals. One helped form the Econometric
Society and edited its journal for 21 years, and defined the problems that
make economics and especially finance distinct from the statistics of biol-
ogy. Ragnar Frisch is still considered a father of modern econometrics.
It was his Norwegian colleague and prodigy, Trygve Haavelmo, though,
who almost single-handedly inspired the agenda which the Cowles
Commission would adopt to establish a discipline distinct from math-
ematical statistics.
Trygve Haavelmo and Ragnar Frisch each won Nobel Memorial
Prizes for their contributions. The committee recognized these individ-
uals, first in 1969, in the first year of the award, and then two decades
later, in 1989, for contributions that may have been difficult to under-
stand and absorb at the time, but which were eventually fully incor-
porated into our toolbox of methods to properly employ data in our
financial models.
The story does not end with the Cowles Commission of the 1940s,
though. From there, Cowles members such as the Great Minds Jacob
Marschak, Harry Markowitz and Kenneth Arrow took the results of
the Cowles Commission of the 1940s and created the plethora of
tools now employed in quantitative finance. These subsequent inno-
vations were documented in the book in this series entitled The Rise
of the Quants. We now rely on the mean and variance approach that
arose first from the explorations of Carl Friedrich Gauss, a remarkable
Great Mind who also spurred the creation of modern statistics and
regression modeling.
The works of these individuals have defined finance theory ever since.
Their contributions allow us to summarize data, construct and then test
financial models, and have allowed both practitioners and policy makers
to digest huge amounts of data to enhance the efficiency of our modern
markets and economies. The modern-day products of their contributions
are the test of the genius of the Great Minds described here.
Glossary
Ordinary Least Squares A methodology that fits a straight line through a number
of points to minimize the sum of the squares of the distances from the best
fit lines to the points.
P-value Also called the calculated probability, is the probability of finding the
observed results when the null hypothesis H0 is true.
Polar Coordinates The description of points on a Cartesian or Complex plane
based on their distance and orientation from the origin.
Polygon A multisided geometric shape with at least three sides and angles.
Probability Density Function A function of a continuous random variable that can
be integrated over an arbitrary interval to determine the probability that the
value of the variable falls within the interval.
Random Walk The expectation that a security return at time t is equal to its last
period value plus a stochastic (random) component that is an independent
and identically distributed with zero mean and variance 2.
Real Number A number that can represent the distance along the number line.
Rectangular Coordinates The description of points on a Cartesian plane based
on the distance from the origin in the direction of each of the dimensions.
Regression A technique used to fit a dependent variable as a weighted sum of
independent variables.
Regular Polygon A polygon for which each apex is equiangular and each side is
of the same length.
Second Moment A weighted measure of the deviation of a random variable from
its mean or first moment.
Simultaneity A condition that occurs when an independent variable and the
postulated dependent variable influence each other.
Special Theory of RelativityEinsteins theory that specified the relationship
between space and time. It is based on the concepts that physical laws are
identical for all non-accelerating frames of reference, and that the speed of
light in a vacuum is a constant for all observers, regardless of their motion
relative to the light source.
Standard Deviation A measure of the spread of numbers, as calculated by the
square root of their variance.
Static The consideration of mathematical, physical, or economic relationships
that do not change over time.
Students t Describes properties of the mean of small samples from a normal
distribution when the population standard deviation is unknown.
Taylors Series The expression of the range of a function arising from deviations
of its domain as represented by an infinite series of the functions derivatives
and the deviations of its domain.
Glossary 255
Type-I Error The incorrect rejection of a correct null hypothesis. This rejection is
often called a false positive.
Type-II Error The failure to reject a false null hypothesis. This acceptance is often
called a false negative.
Unit Circle A circle centered at the origin of orthogonal axes and with a radius
of unit length.
Variance A specific measure of the sum of distances of a set of data points around
their mean value. These distances can be calculated by summing the square of
the coordinates for each point.
Volatility A measure of the degree of uncertainty and unexplained movements
of a variable over time.
Index1
A C
Adrian, Robert, 46 capital asset pricing model
Aitken, Alexander Craig, 134, (CAPM), 79
137n2 Cardano, Girolamo, 20, 21
Al-Khwarizmi, 19 Cayley, Arthur, 134, 137n1
asympotic, 1679 center of mass, 40, 49, 54, 149
Chebyshev, Pafnuty, 56
Classical Model, 251
B Clifford, William, 89, 90
Bell, Eric, 157, 158 collinearity, 229
Bernoulli, Jacob, 47, 49, 50, 56, 65, Columbus, Christopher, 38
91 complex analysis, 30, 64
Born, Max, 12 complex number(s), 1731, 41
Boscovich, Roger, 37 Comte, Isadore, 76
Brahe, Tycho, 48 constructable polygon, 16, 29
1
Note: Page number followed by n denote footnotes
H L
Haavelmo, Trygve, vii, 196, 197, Lagrange, Joseph-Louis, 9, 30, 31
220, 221, 22339, 2413, Lambert, Johann, 42
247, 249 least squares, vi, 9, 11, 23, 26, 34,
Heisenberg, Werner, 12, 92 37, 406, 50, 53, 57, 59,
heteroscedasticity, 213 61, 646, 76, 79, 95, 97,
Hilbert, David, 12, 139 104, 1337, 147, 165, 175,
Hitler, Adolf, 12, 117, 144, 161, 191, 197, 213, 214, 216,
162, 196 231, 232, 237, 248
Hotelling, Harold, vi, vii, 97, 109, Legendre, Adrien-Marie, 30, 435,
137, 14863, 1659, 53, 56n3, 57, 59, 64
1713, 175, 192, 196, 205, Leibniz, Gottfried, 44
208, 221, 248 Lexis, Wilhelm, 94
Hudson, Henry, 154 Lorenz, Ludvig, 17
Humboldt, Friedrich, 61, 62 Lyapunov, Aleksandr, 56
I M
identification problem, 229, 233, Malthus, Thomas, 77
238n3 Markov, Andrey, 56
imaginary number, 13, 18, 20, 224, Marschak, Jacob, vii, 196, 197,
27, 30 21921, 228, 229, 233, 249
independent and identically Maxwell, James, 61, 86, 87
distributed, 46, 54, 237 mean(s), vi, ix, 4, 8, 11, 12, 1416,
intertemporal, xii 19, 21, 23, 30, 36, 37,
3941, 45, 46, 48, 504,
56, 57, 72, 75, 79, 91, 94,
J 95, 97, 100, 123, 127, 129,
Johnson, Alvin, 196 131, 132, 1357, 142, 146,
147, 161, 166, 209, 214,
215, 221, 228, 231, 238n1,
K 248, 249
Kepler, Johannes, 38 Mendel, Gregor, 116
Keynes, John Maynard, 117, 211, Michelson, Albert, 17
212, 231, 232 Mill, John Stuart, 100, 107
Kohlrausch, Rudolf, 61 Modigliani, Franco, 196, 198, 233
Koopmans, Tjalling, 197, 232, 233, Moivre, Abraham de, 22, 24, 28, 46,
238n3 48, 50, 56, 91
260 Index
N R
Newton, Isaac, 4, 9, 22, 25, 39, 44 Ramsey, Frank, 171, 172
normal distribution, vi, 45, 4656, real number, 17, 22, 24, 50
76, 90, 91, 936, 123, 125, Rectangular coordinates, 254
129, 166, 195, 213, 247 regression(s), xi, 46, 76, 79, 80,
number theory, 13, 30, 41, 157, 172 957, 104, 125, 1315,
137, 191, 192, 194,
21317, 217n3, 231, 233,
O 237, 238n1, 249
Occam's Razor, 39 regular polygon(s), 10, 13, 15, 16,
Ockham, William of, 39 27, 28, 33
Olbers, Heinrich, 34, 42 Ricardo, David, 70, 71
Oppenheimer, Robert, 12 Riemann, Bernhard, 12
ordinary least squares, 79, 214 Roos, Charles, 195, 2068, 219
P S
Pascal, Blaise, 468 Samuelson, Paul, 194
Pauli, Wolfgang, 12 Savage, Jimmie, 146
Pearson, Egon, 103, 104, 126, 127, Schumpeter, Joseph, 207, 208,
140, 141, 143, 226, 227 211, 220
Pearson, Karl, vi, 80, 827, 8997, Sharpe, William, 228
99107, 116, 118, 1202, Simon, Herbert, 73, 77, 198, 233
125, 140, 141, 160, 226, simultaneity, 229, 237, 241
248 Snedecor, George Waddel, 142
Piazza, Guiseppe, 33, 34 Special Theory of Relativity, 107
Pisa, Leonardo da, 19 standard deviation, 50, 80, 93, 94,
Planck, Max, 12 104, 121, 124, 166
polar coordinates, 26 static(s), 92, 196, 206, 212,
polygon(s), 10, 1316, 279, 33, 59 217n1
probability density function, 40, Stratton, J.M., 118, 121, 123, 130,
45, 53 160
P-value(s), 104 Student's t, 132, 142, 160
Index 261
T W
Tartaglia, Niccolo, 19, 20 Wald, Abraham, 161, 169
Turnbull, Herbert Westren, 134, 137n2 Wallis, John, 21, 22
type-I error, 104 Waugh, Frederick, 211, 213, 214,
type-II error, 104 217, 220
Weber, Wilhelm, 602
Weldon, Walter, 92, 93
U Wessel, Caspar, 22, 23, 26, 225
unit circle, 13, 15, 268 Williams, John Burr, 165, 179
V Y
variance(s), vi, 46, 525, 91, 93, 94, Yntema, Theodore, 219, 220
96, 97, 1214, 127, 12931, Yule, G.Udny, 126
135, 136, 140, 142, 147,
166, 167, 189, 228, 249
Veblen, Oswald, 158, 167 Z
Veblen, Thorstein, 158 Zach, Franz von, 34
Verrezzano, Giovanni, 38 Zimmerman, Duke Carl Wilhelm,
von Neumann, John, 12, 139, 171 8, 9