|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
ESSAYS ON TECHNOLOGICAL CHANGE
KEVIN W. CHRISTENSEN
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
Kevin W. Christensen
To my family. Without their love and support none of this would have been possible.
I would like to thank my advisors, Elias Dinopoulos and Chunrong Ai, for their
help with my research. James Seale, Jr. and Doug Waldo were also helpful in developing
this finished product. David Figlio, Sarah Hamersma, Jonathan Hamilton, Larry Kenny,
and other professors in the Economics Department were generous with their time and
input. Special thanks go to committee member and graduate coordinator Steven Slutsky.
Professor Slutsky's insights, guidance, and advice were valuable at all stages of my
The successful completion of my dissertation would not have been possible without
my family and friends. My parents, Homer and Charlene Christensen, my sister Michele
Bach-Hansen, and her husband Scott, were unwavering in their love and support. My
nieces, Madison and Kayla, provided me with much needed distraction, entertainment,
and joy. Friends in Virginia, Florida, and elsewhere were always available when I
needed them. Finally, I am very grateful to Burgin Uinel. She believed in me when I was
sure no one else did.
TABLE OF CONTENTS
ACKNOWLEDGMENT S .............. .................... iv
LI ST OF T ABLE S ................. ................. vii........ ....
LIST OF FIGURES .............. .................... ix
AB S TRAC T ..... ._ ................. ............_........x
1 INTRODUCTION ................. ...............1.......... ......
2 A MODEL OF ENTREPRENEURSHIP AND SCALE-INVARIANT GROWTH....4
2.1 Introduction ......._................ ...............4. ....
2.2 Previous Literature .................. ............ ...............5......
2.2.1 Endogenous Growth Literature ................. ............_ ........ .......5
2.2.2 Finance and Growth Literature: Theory .............. .....................
2.3 The M odel ................. ......... .. ...............9......
2.3.1 Consumer Utility ................. ......... ... ...............9. ....
2.3.2 Competition, Prices, and Profits ..............._ .............._ ........ ...11
2.3.3 Innovation ..............._ ................. 13........ ....
2.3.4 Finance Sector ................. ...............15......... ....
2.3.5 Financial Intermediation ......... .............._ ........_ ............1
2.3.6 Stock M market ................._ .......... ...............17.....
2.3.7 Financial Sector Equilibrium ..............._ ...............18 ........._....
2.3.8 Labor M market .............. ...............18....
2.4 Balanced Growth Equilibrium ..............._ ........ ........_ ............1
2.4.1 Transitional Dynamics ..............._ ...............20 ......... ....
2.4.2 Economic Growth ..............._ ...............22 ......... ....
2.4.3 Comparative Statics .............. ...............23....
2.5 Conclusions and Extensions .............. ...............25....
3 THE EFFECT OF PRUDENT INVESTOR LAWS ON INNOVATION .................30
3.1 Introduction ................. ...............30........... ....
3.2 Background ................. ......... ...............32.......
3.2.1 Prudence of Investment ................. ...............32................
3.2.2 Previous Literature ................. ...............34........... ....
3.3 Data and Empirical Methodology ................. ...............36........... ...
3.3.1 D ata ................... ............ ...............37.......
3.3.2 Empirical Methodology .............. ...............39...
3.4 Tests of Exogeneity & Benchmark Regressions ................. ......................42
3.4.1 State Innovative Output and the Timing of Adoption ................... ........42
3.4.2 Evidence from a Long Difference............... ...............4
3.5 Prudent Investor Laws & Innovation............... ...............4
3.5.1 Indirect Investments: Venture Capital .............. .....................4
3.5.2 Direct Investments: R&D Expenditures .............. ....................4
3.5.3 Alternative Mechanisms .............. ...............48....
3.6 Conclusion .............. ...............49....
4 DO PATENT ATTORNEY S MATTER? ............. ...............64.....
4.1 Introduction ................. ...............64........_. ....
4.2 Literature Review ...._._. ................. ......._._. .........6
4.2.1 Theoretical Models............... ...............65.
4.2.2 Previous Empirical Analyses ................. ..............................67
4.3 Sources and Descriptive Statistics .............. ...............69....
4.3.1 Data Sources .............. ...............69....
4.3.2 Description of Variables ........._..._ ...._.... ...._.__ ................ 71
4.4 Empirical Methodology .............. ...............77....
4.4.1 Regression Specifications .............. ...............77....
4.4.2 Endogeneity of Lawyer Choice ................. ..............................78
4.5 Re sults ................. ............ .................8 1....
4.5.1 Estimated Impact of Lawyers ................. ............. ......... .......81
4.5.2 Examiner Experience and Generality .............. ..... ............... 8
4.5.3 Examiner Effects ................... ............ ...............84......
4.5.4 Experience as a Proxy for Quality .............. ...............85....
4.6 Conclusion .............. ...............86....
5 CONCLUSION................ ..............12
A PROOFS OF PROPOSITIONS ................. ...............125...............
B THE BLUNDELL-B OND ESTIMATOR ........_.......... __. ............... 128 ....
C INSTRUMENTAL VARIABLES AND THE ENDOGENEITY OF LAWYER
CHARACTERI ST IC S ............_...... ..............1 2.....
REFERENCE S .............. ...............137....
BIOGRAPHICAL SKETCH ............_...... ...............143...
LIST OF TABLES
2- 1. C omparative Stati c s................. ...............29...._.__ ..
3-1. Correlation Matrix ..........._.._ ....... ...............51...
3-2. Descriptive Statistics .............. ...............52....
3-3. Descriptive Statistics by Year............... ...............53..
3-4. Year of Adoption of UPIA (or equivalent) ...._ ......_____ ...... ......__......5
3-5. Comparison of Adopters and Non Adopters ...........__...... ..........__......55
3-6. The Timing of Adoption ...........__..... .___ ...............57..
3 -7. Impact of Prudent Investor Laws over Long Difference.............__ ..........___.....5 8
3 -8. Estimates of the Impact on Venture Capital Investments in a State ........................59
3 -9. Estimates of the Impact on R&D Expenditures in a State ................. ................ ...60
3 -10. Estimates of the Impact on Citation Weighted Patent Counts in a State ........._.......62
4-1. Variable Descriptions .............. ...............88....
4-2. Technology Subcategories Descriptions .............. ...............89....
4-3. Count of Unique Occurrences, by Subcategory, When Identification of First or
Most Experienced Lawyer is Known ................. ...............90........... ...
4-4. Averages by Subcategory, When Identification of First or Most Experienced
Lawyer is Known .............. ...............91....
4-5. Number of Patents by Country and Subcategory, When Identification of First or
Most Experienced Lawyer is known ................. ...............93........... ...
4-6. Correlation between Grant Lag and Independent Variables, by Technology
Sub category ................. ...............94.......... ......
4-7. Impact of Representation by Patent Attorney or Agent .............. ....................9
4-8. Estimating Grant Lag, Without Lawyer (Patents Where First Lawyer is Known) .103
4-9. Impact of Lawyer Experience, Using First Lawyer Listed ................ ..................106
4-10. Estimating Grant Lag, Without Lawyer (Patents Where Most Experienced
Lawyer is Known) ................. ...............109................
4-1 1. Impact of Lawyer Experience, Most Experienced Lawyer Listed ................... .....1 12
4-12. Impact of Examiner Experience, Controlling for First Listed Lawyer. .................1 15
4-14. Predicted Impact of Lawyer Quality, First Lawyer ................. ............ .........121
4-15. Predicted Impact of Lawyer Quality, Most Experienced Lawyer ................... ...... 122
C-1. OED Examination Dates and Passing Rates............... ...............136.
LIST OF FIGURES
2-1. Equilibrium Conditions for Model .............. ...............27....
2-2. Stability of Balanced-Growth Equilibrium............... ..............2
C-1. Number of Utility Patents Granted, Annually, by the USPTO .............. ..............13 5
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
ESSAYS ON TECHNOLOGICAL CHANGE
Kevin W. Christensen
Chair: Elias Dinopoulos
Cochair: Chunrong Ai
Major Department: Economics
My dissertation consists of three essays on the economics of technological change.
The first essay develops a theoretical model that describes how financial intermediaries
may influence economic growth. Previous theoretical models on the topic predict that
economies with larger populations will grow at faster rates, something which has not
been empirically supported. This paper corrects this "scale effects" issue by extending an
existing model of economic growth, without scale effects, to include a Einance sector.
The Einancial intermediary evaluates potential entrepreneurs and their ex ante potential
for being an entrepreneur. Upon receiving a positive rating from the intermediary, the
entrepreneur receives money for R&D which, in turn, may lead to successful innovation.
Changes in the steady-state growth rate are explained by shifts in parameter values.
In a spirit similar to the first essay, the second essay considers the impact the
adoption of prudent investor laws had on innovation. These laws were primarily adopted
by states in the late 1990's and expanded the scope of investment options available to
financial intermediaries to include new and untried enterprises and venture capital.
Various specifications using state-by-industry patent counts, venture capital
disbursements by state, and R&D expenditures were used to test whether these laws
affected technological change. The empirical results show that, contrary to previous
evidence, prudent investor laws had only a small effect on technological change. This
suggests that the impact of financial intermediaries on economic growth may be bounded.
The final essay explores the role that another intermediary has on technological
change. As active participants in the patenting process, patent attorneys are involved in
writing and defending the claims on an application (among other things) and thus can
help to establish the scope of patent protection. This chapter explores the value added of
patent attorneys by looking at how more experienced lawyers affect the time between
filing of an application and the date a patent is granted. It has been found that more
experienced attorneys can reduce the grant lag, but the reduction depends on the
invention's technology. This research is the first that considers the role of attorneys in
the patent process.
Technological change affects every sub-discipline of economics. Micro-
economists may explore the role of research and development in competition. They
might also explore the role that technology plays in determining industry composition.
Labor economists may be concerned with increased worker productivity as a result of
new machinery and equipment. Public economists may consider the role of the Internet
in increasing test scores among minorities and the poor. Research on international trade
may estimate the impact and flow of international technology spillovers. The
overarching theme of these scenarios is that technological change is generally good for an
economy.l Nowhere in economics is that made clearer than in the literature on economic
growth. As Schumpeter said,
The fundamental impulse that sets and keeps the capitalist engine in motion comes
from the new consumers' goods, the new methods of production or transportation,
the new markets, the new forms of industrial organization that capitalist enterprise
creates....This kind of competition is...the powerful lever that in the long run
expands output and brings down prices.... (1950, pp. 83-85)
This idea is found in theoretical models, such as Solow (1956) and Romer (1986), which
provide a framework for understanding how advancements in technology can positively
affect economic growth. However, theoretical models require assumptions that abstract
from the real world and can assume away some features to facilitate understanding or
SPolitical economists, such as Karl Marx, may disagree with this assertion.
computation. One oft-ignored element is the role of intermediaries in facilitating
Economic theory provides two primary reasons for the existence of intermediaries:
cost and information. Intermediaries specialize in a particular field and therefore have
capabilities and knowledge that more generalized firms (e.g. a manufacturing company)
may not have. It is not that the firms could not acquire these capabilities but that, by
specializing, the intermediary is better informed and may provide the services cheaper
than a general firm could achieve alone. Given this, use of an intermediary causes
efficiencies that may, in turn, lead to an increased rate of innovation.
This dissertation explores the intersection between technological change and
intermediaries. Specifically it considers the role that of financial and legal intermediaries
have on facilitating technological change. The second chapter of the dissertation presents
a theoretical model that outlines the role of financial intermediaries in fostering economic
growth through technological change. The third chapter empirically tests the effect a law
change affecting the types of investments financial intermediaries could be made.
Combined, these essays show that financial intermediaries can positively affect
technological change but are not necessarily guaranteed to do so since financial
intermediaries are complements to, not substitutes for, other processes such as research
and development or entrepreneurial initiatives. The fourth chapter considers the role of
patent attorneys in the patent approval process. Using a unique dataset on patent
attorneys it is shown that patent attorneys can affect the time in which a patent is
approved. This is in stark contrast to previous empirical and theoretical literature on the
topic which holds the view that a patent examiner works independent of other factors.
The final chapter summarizes the findings from each the previous three chapters.
A MODEL OF ENTREPRENEURSHIP AND SCALE-INVARIANT GROWTH
As early as Schumpeter (1934), the finance sector was proposed as an important
component in the growth process. Bencivenga and Smith (1991), Greenwood and
Javanovic (1990), and King and Levine (1993b) later formalized that proposition within
the context of endogenous growth theory.2,3 Beck and Levine (2004), Benhabib and
Spiegel (2000), and Rajan and Zingales (1998), among others, have empirically shown
that the finance sector plays an important role in fostering economic growth. They
continue a long line of empirical papers evaluating the relationship (see Levine (2004) for
a review of empirical and theoretical papers). Compared to empirical research,
theoretical work on the finance-growth hypothesis has slowed. As a result, some
innovations in endogenous growth theory remain outside the finance-growth literature.
One of the most significant omissions is the treatment of population as a variable
changing over time rather than as a parameter.
Earlier models of endogenous growth incorporated the undesirable property of
scale effects. These models predict that as population increases, the long-run rate of
growth also increases, implying, ceterus paribus, larger economies grow at faster rates.
For some time, this was considered to be a strength to the theory as growth in population
2 Other, more classical references in the literature are Goldsmith (1969), McKinnon (1973) and Shaw
3 Romer (1986), Grossman and Helpman (1991), and Aghion and Howitt (1992) are major contributors to
the endogenous growth literature.
was deemed analogous to globalization. However, time series tests by Jones (1995a,
1995b) showed that growth had remained roughly constant regardless of scale of
population, contradicting these models. As a result, theorists began to develop second-
generation models of endogenous growth that had growing population. In spite of this
innovation, no previous paper modeling the relationship between Einance and growth has
been updated to account for the scale effects issue and a disconnect remains between the
Einance-growth theories and the most state of the art endogenous growth models.
This chapter attempts to bridge that gap by combining a second-generation model
of endogenous growth with the King and Levine (1993b) Einance sector. The general
equilibrium model presented here includes growing population and is shown to have a
balanced growth equilibrium that is saddlepath stable. Fluctuations in parameter values
explain changes in the growth rate of the economy. The rest of the chapter proceeds as
follows. The next section of the chapter reviews the relevant literature. The third section
introduces the equilibrium conditions for consumers, producers, and Einancial
intermediaries. These conditions are used to specify the balanced growth, transitional
dynamics, and comparative statics presented in the fourth section. The Einal section
offers conclusions, limitations, and proposes extensions for the model.
2.2 Previous Literature
2.2.1 Endogenous Growth Literature
In the late 1990's three papers, Young (1998), Howitt (1999), and Segerstrom
(1998) were published as the core of the second generation endogenous growth models,
each with a distinct answer to scale effects. Young introduced the idea that both
horizontal and vertical product competition offsets the scale effects problem. More firms
producing at the same level of quality but with different varieties will reduce the spoils
available to any one producer. As a result, growth does not reach the high levels it did in
first-generation models. Howitt translated Young's original idea into a more traditional
Schumpetarian growth model. In doing so, he reintroduced the result that R&D subsidies
provide a positive impact on growth which was lacking in Young's original model.
Segerstrom's model used a quality ladder approach and an R&D difficulty index to offset
the impact of larger population size. The difficulty index removes the population scale
effects while at the same time explaining why R&D employment has increased without a
commensurate increase in innovation. Neither of the other two models explains this
phenomenon. Segerstrom's model also allows for a positive impact of R&D subsidies on
2.2.2 Finance and Growth Literature: Theory
Theoretical models on the finance-growth relationship are varied in their scope and
use of endogenous growth fundamentals. Bencivenga and Smith (1991) consider how a
developing finance sector alters the composition of consumer savings using a three period
overlapping generations model. As in other models, the introduction of a finance sector
increases the accumulation of capital. It is shown that these changes do not occur as a
result of changes in savings behavior but instead are a direct result of the intermediary
efficiently allocating consumer savings. Greenwood and Javanovic (1990) also consider
an evolving finance sector with endogenous improvement in production inputs. In their
model, the finance sector matures as the income of the population increases. Higher rates
of savings drive the finance sector (and economy) forward in the development process.
As the finance sector evolves, the rate of return on capital increases. This increased
return is what drives growth in the economy. Therefore a country with a more mature
finance sector would have a higher level of growth than a relatively less-mature
economy. Each of these models utilizes an AK endogenous growth model as a starting
point where capital accumulation is determined endogenously.
Unlike the previous two papers, King and Levine (1993b) use Aghion and Howitt' s
(1992) endogenous growth model as its foundation and does not consider an evolving
finance sector. Instead, the maturity of the finance sector is treated as given and its
impact on the introduction of intermediate products is evaluated. The financial
intermediary acts as a filter of prospective entrepreneurs that seek financing. Only
projects presented by skillful entrepreneurs will be able to obtain funding. Those
individuals without a positive rating cannot compete for the next innovation and instead
become production workers. Another change from the previously mentioned models is
the introduction of a stock market which accumulates consumer savings and provides
revenue to fund entrepreneurial ventures through the initial offering of stocks.
A paper by Morales (2003) is the most recent known paper modeling the finance-
growth relationship.4 It is based on Howitt and Aghion's study (1998) which
incorporates a "leading-edge technology parameter" that provides a similar function to
Segerstrom's (1998) R&D difficulty index and uses both capital and labor as factors of
production. Two elements make their paper unique from other finance-growth models.
First, the model introduces capital as an input in production. By including capital, both
capital accumulation and technological change lead to economic growth. The second
element is the inclusion of moral hazard between the financial intermediary and the
researchers. In spite of the presence of moral hazard, her results show a positive
SAghion, Angeletos, Banerjee, and Manova (21 *4) considers a related (but not identical) issue of how
volatility affects technological change. The model developed considers the investments of finitely lived
entrepreneurs but assumes the number of entrepreneurs is constant over time.
relationship between the finance sector and the success rate of proj ects and research.
While the addition of capital and moral hazard are significant contributions, the scale
effects issue remains. Therefore the steady-state analysis presented in her paper is only
stable for a set population not for one growing over time.
This chapter presents a model of growth without scale effects by combining King
and Levine's (1993b) finance sector with Segerstrom's (1998) model to obtain a general
equilibrium model that has a balanced growth equilibrium and is saddlepath stable.
Similar to Morales (2003), the model shows a positive relationship between finance and
economic growth spurred through technological change. Unlike Morales' work, labor is
the only input and any potential moral hazard is assumed away. To combine the King
and Levine and Segerstrom models, several changes have been made. In the models
mentioned above, technological advancement improves intermediate goods that are used
to produce a single consumable product. The model presented in here utilizes product,
rather than process innovations. That is, rather than multiple inputs for one final product
there are multiple final goods. Entrepreneurs compete to innovate to the next quality
level of the final good in a specific industry.' Economic growth is observed through
increases in consumer utility which is affected by the quality and the quantity of the
goods consumed. The second important change is the endogenous treatment of
entrepreneurial competitors. This is done to reflect how the growing population (and
increased consumer demand) affects the number of entrepreneurs competing for the next
SOne advantage of this structure is that it fits well with the empirical observations of Hellman and Puri
(2000) where a venture capitalist is likely to invest in a technology that is pushing out the technological
frontier as opposed to one creating horizontal product innovation. The predicted direction of this model
(but not necessarily the magnitude) can help in understanding the role venture capital plays in economic
innovation. Finally, only a portion of R&D employment directly affects the innovation
rate whereas Segerstrom's model attributes all non-production workers as R&D labor.
Since the model introduces a Einancial sector, labor allocated to the financial intermediary
is not growth promoting and does not per se impact the innovation probability. However
their importance to the growth process will be highlighted later in the chapter.
2.3 The Model
The description of the model begins with a discussion of consumer preferences.
Once the consumer equilibrium condition is established the producer side, innovation
process, and the role of Einancial intermediaries are developed. This section concludes by
elaborating on the labor market. These market equilibrium conditions will be used in the
fourth section to estimate the balanced growth equilibrium values of per capital
consumption and per capital R&D difficulty.
2.3.1 Consumer Utility
The model uses dynastic families as outlined by Barro and Sala-i-Martin (2001)
and used by Segerstrom (1998). Dynastic families choose to maximize the utility of all
family members over an infinite horizon. That is, current family members are altruistic
towards their current and future relatives and make consumption choices with them in
mind. By using the dynastic family assumption the model bypasses the problems of
finitely lived people and allows for a single and unified utility function to be maximized.
Assuming that each individual has an identical utility function, the utility equation is the
product of the individual discounted utility and the population for the entire economy
summed over an infinite horizon.
Each individual in the economy has a discounted utility function equal to
Se t In[u(t)]dt where p is the discount factor and u(t) is the subutility. Population at
time t is N(t) = e"' when initial population is normalized to 1 and the exogenous growth
rate of population is n (births minus deaths). For optimization purposes p is assumed to
be greater than n.6 The simplified product of these components over an infinite horizon is
U = e "" InBu(t)]dt (2-1)
Product quality and consumer demand are introduced in the subutility. Quality
levels are sequential so an industry cannot produce the j+1 quality product without the j
quality product already having been discovered. Each industry can produce goods of
different qualities at the same time. Once price accounts for quality differences, each
product within the same industry substitutes perfectly. The subutility function is defined
Inu o)] In I d(j, m, t) m. (2-2)
The quality of product j is denoted by A' where the parameter Ai represents the step
size of innovation. As Ai increases, the difference between the quality of the new good
and the old good increases. Since product quality improves with each innovation, Ai must
be strictly greater than 1. Quantity demanded by an individual consumer is denoted by
dcj, mi, t) for a particular quality (j) and industry (0i) at a point in time (t). The total affect
6 See Barro and Sala-i-Martin (2001, p. 67) for a thorough explanation of this restriction and the
of consumption on utility is simply the product of the quality and demand summed across
all industries, which are indexed along a continuum from 0 to 1.
At every point in time consumers choose the amount to spend on an industry's
product. Given a unitary elasticity of substitution between goods of differing qualities,
per capital demands at a point in time are d = c per capital consumption divided by the
price of the good.' To break ties, it is assumed that the consumer purchases the more
advanced quality product. Consumers only choose c(t) and treat prices and qualities as
given so over time, per capital consumption may vary. Taking this into account and
substituting the demands as noted above into the subutility function, maximizing (2-3) is
equivalent to maximizing (2-1).
ea r".' In c(t)dt (2-3)
The family's optimal consumption is bounded by the growth of per capital assets,
ci(t). Consumer assets change due to wages w(t), stock market dividends r(t)a(t),
consumption c(t), and division of assets among new family members na(t). Therefore,
the constraint for the maximization of utility above is ci(t) = w + r(t)a(t) c(t) na(t) .
Solving the dynamic constrained maximization problem yields
= r(t) p. (2-4)
2.3.2 Competition, Prices, and Profits
Consider the only possible competitive case where there are two firms in an
industry each producing different qualities, j and j+1. The producer of the cutting edge
SLi (2003) extends Segerstrom's (1998) model to account for non-unitary elasticities of substitution.
technology, j+1, is called the quality leader and the other firm is referred to as the quality
follower. Consumers are indifferent between the qualities if the effect on utility is the
same for either good. That is if, ii d( j, mi, t) = il""d( j + 1, mi, t) Recall that demands are
equal to per capital consumption divided by price and that consumers allocate the amount
of consumption to an industry, not a specific quality. Given this, the equivalent price
indifference equation is p~,, = Alp, Assuming Bertrand competition prevails in all
industries, the quality follower sets its price at the lowest possible level, the marginal cost
of production. Since labor is the only input and one unit of labor is required to produce
one unit of output the price of the quality leader is p, = lw This is the case for all
quality leaders regardless of industry. Assuming consumers prefer the quality leader's
product when formally indifferent, the quality leader is the sole producer in equilibrium
given the contestable market. Since this will be true for all industries and qualities, the
prevailing market price for the economy is
p = Alw (2-5)
The profits of the quality leader are equal to the price-cost margin of each product
times the number of products sold. Since consumers are assumed to have identical
utilities the demands for each individual are also the same, implying that market demand
for a specific time, quality and industry is DOi, mi, t)=N(t)dyi, mi, t). The profit equation
for the sole producer may therefore be simplified to
Fot)= Ntc*,t) (2-6)
The profits earned by the quality leader are greater than zero by definition of N(t),
c(t), w, and ii. It is the desire for these profits that leads to innovation.
Each innovation attempt may advance only one step beyond the current quality
level and successful innovation is far from certain. Each attempt is governed by a
Poisson process where the probability of innovation increases with the amount of labor
used in R&D. In this model there are two components of R&D labor: the researchers
and the Einancial intermediary's employees. Unlike in Segerstrom's (1998) model, not all
R&D employees affect the rate of growth of innovation. The labor used by the Einancial
intermediary does not directly advance research so only researcher's labor, e, increases
the probability of success. It is possible that multiple entrepreneurs in the same industry
may be positively rated by the intermediary so that there is more than one competitor for
the next quality step. The endogenous variable H~im, t) represents the number of
entrepreneurs that attempt to innovate in that industry at each point in time. Even though
Financial intermediary employees and R&D workers are represented by parameters, the
number of competitors and therefore the total number of employees will grow over time.
It is plausible to think that early stage advancements are easier than later stage
advancements. That is, simpler innovations take no time whereas more complex
innovations require extensive testing, or perhaps even a lengthier review process by
government agencies. As time passes and the industry moves up the quality ladder the
probability of successfully innovating decreases. To account for this, the innovation
probability uses an industry specific R&D difficulty index, X~im, t), which increases over
time but affects the innovation probability negatively. In spite of the growth in
competitors it is possible that innovation may stay constant or decrease depending on if it
is dominated by the R&D difficulty index. The probability that an industry innovates to
the next quality level is
O(, t) = H(m, t), (2-7)
where A is a productivity parameter.
Assumption 2-1: The R&D difficulty index increases at a rate equal to
= pe(mi, t), where the parameter puE (0,1]. This implies that complexity of
products rises as firms become more innovative.
There are no spillovers across quality levels. A veteran participant in the jth patent
race now competing in the j+1~st race has no advantage over a relative newcomer. Any
participants in the patent race for the jth quality must start from the beginning of the
process to reach the j+1~st quality level so there are no spillovers between previous and
current research nor is there any spillover between researchers in the same patent race.
Thus, the industry innovation rate is the product of individual competitor probabilities,
(m,t), multiplied by the number of competitors so that, O(mi, t) = #(mi, t)H(mi, t) .
Each attempt requires a different request for startup capital by entrepreneurs from
investors. The return on investment to these investors is the expected profits from the
sale of the product. Given the competitive makeup of all industries, the profits associated
with one quality level disappear when the next innovation occurs. If the current industry
quality leader attempts to innovate twice to advance two steps up the quality ladder, the
leader becomes indebted to two cohorts of investors. Further, by innovating to the j+1
quality, the firm eliminates the demand for its j quality product and thus cuts off revenues
from that product and dilutes the shares contrary to the interest of its original set of
investors (a similar concept to Myers and Majluf (1984)). This business stealing outcome
and the inability to repay two cohorts of investors are the reasons each entrepreneur will
only choose to advance one quality rung at a time. Therefore, while possible, it is not
desirable for an entrepreneur to attempt two successive levels of innovation. Further
advances in quality must come from outside the firm.
2.3.4 Finance Sector
As mentioned previously, profits are the incentive for innovation, but there are
steps that must be taken before these profits are realized. The Einance sector is composed
of two related areas, a financial intermediary and a stock market. The symbiotic
relationship between the two areas is critical to technological change and thus economic
growth. In the model, the intermediary provides a means of assurance to investors by
rating each entrepreneur and entrepreneurial venture as either good or bad. Upon a
positive rating, the startup capital necessary to participate in a patent race is provided
through the stock market. Only positively rated firms will receive startup capital and be
able to attempt to innovate.
In the real world, an intermediary provides more than just a rating. Startup capital
to an entrepreneurial proj ect is invested with the expectation that there will be a return on
that investment. During the time between investment and realization, the intermediary
firm may provide strategic advice, monitoring, or lower the learning curve for new
entrepreneurs (in the context of venture capital, see Hellmann and Puri (2000, p. 960).
Finally, by investing in a company, a financial intermediary firm sends a signal to future
investors that the project, while risky, has potential. The model presented here eliminates
the financial and mentoring responsibilities of the intermediary and focuses solely on the
signaling aspect. However, the productivity parameter in the innovation probability
could be interpreted as the value added impact from mentoring. The sole proactive
responsibility of the intermediary in the model is to provide assurance to stock market
investors. The rating guarantees that the proj ect can succeed but does not guarantee that
it will be the first to succeed.
2.3.5 Financial Intermediation
It is assumed that individuals posses traits that will make them successful with a
probability a. The intermediary can reveal a potential entrepreneur's ability, with
certainty, by investigating the individual at a cost of units of labor. In equilibrium, the
maximum value an intermediary is willing to invest on a rating for an individual proj ect
is the expected value of the proposed entrepreneurial project. The structure of the model
is such that each equivalent quality step results in the same amount of profit regardless of
industry. It is possible that multiple entrepreneurs in the same industry may be positively
rated by the intermediary so that there are multiple competitors for the next quality step;
however, each potential entrepreneur is considered on a case by case basis. With q
representing the expected discounted value of the entrepreneurial venture, the equilibrium
conditions for a financial intermediary are
aq = wf (2-8)
q = #(co~, t)py(t) we (2-9)
With the perfectly competitive labor market w is the same wage as in the
production side of the model. The stock market value of a firm is represented by v(t).
Proposition 2-1: The expression for equilibrium of one firm, q, is equivalent to
the industry equilibrium condition. See Appendix A for proof.
Combining these equations and solving for v(t) yields the financial intermediary
v(t)= w( +e (2-10
The structure of the model is such that each equivalent quality step results in the
same amount of profit during the same time period, regardless of industry. The financial
intermediary has no incentive to prefer some industries to others since profits are the
same across industries. Due to the symmetric nature of the financial intermediary,
profits, and price equilibria, the rest of the model focuses on the general case where the
innovation rate and entrepreneurial competition is the same across all industries. As a
result, the industry component of all functions from this point on is dropped.
2.3.6 Stock Market
After being rated, an entrepreneur may seek funding via the stock market to start a
new business. This funding is used to pay for R&D that will hopefully lead to an
innovation. The securities issued for new firms compete with those from other industries
and with stocks from already established quality leaders. When making her investment
choices, a rational consumer will make comparisons to a perfectly riskless asset with a
rate of return r(t)dt for a time segment dt. In equilibrium, the expected rate of return for
new stock must be equal to the rate of return on the riskless asset. The expected stock
value of the new firm is equal to the realized dividends plus the expected capital gains for
the time segment dt. The expected value is adjusted downward since the future value
disappears when the next product innovation occurs. The equilibrium condition for the
O(t)dt)dt O(t)dt = r(t)dt .
time segment dtf is therefore d~t) + f)(1_
Taking the limit as dt approaches zero it follo~
r(t) + (t)-
As in Segerstrom (1998), the growth rate of the stock market value of monopoly
profits must be equal to the growth rate of the R&D difficulty index,
One implication of the stock market equilibrium is that as R&D difficulty
increases, the stock market value increases which corresponds with more investment.
This model has a decreasing per dollar impact of financial capital on innovation as time
progresses so that more capital is needed over time to keep innovation probability the
2.3.7 Financial Sector Equilibrium
When both the stock market and intermediary are in equilibrium the entire finance
sector is in equilibrium. Recalling Assumption 2-1, the R&D equilibrium condition may
now be solved. Where x(t), which equals, X(t) N(t), is per capital R&D difficulty.
wx(t) f Aww
+ e = .(2-12)
Aep a [rt) + O(t)(1- p)]l
2.3.8 Labor Market
Employees have two choices of employment; they may work either in the
manufacturing or R&D sectors. Since the wages in these two sectors are the same,
SSee Appendix A for proof.
workers are indifferent between these two j obs. Given full employment, N(t) is the sum
of manufacturing labor (N"id(t)) and R&D labor (NRDt), which includes financial
intermediary labor). The manufacturing labor is equal to the market demand summed
across the total number of industries since it was assumed that each unit of labor supplies
one unit of output.
N~()=j c(t)N(t)il c(t)N(t)il
N" (t = d (2-13)
On a per proj ect basis, entrepreneurial employment can be found in the financial
intermediary condition. Multiplying this by the number of competing entrepreneurial
firms in each industry and summing across all industries yields
NR (t = a + e H(t 5e + e -H(t). (2-14)
Given full employment, the resource constraint for the economy is equivalent to
1= +x~t)- +e, ( 1~ (2-15)
Av a Ae
2.4 Balanced Growth Equilibrium
Now consider the balanced growth equilibrium where all endogenous variables
grow at a constant but not necessarily identical rate. Using (2-7) the balanced growth
innovation rate 0 is
e = H~t)(2-16)
Proposition 2-2: In the balanced growth equilibrium H(t) See Appendix A
It is intuitive that the number of competitors should grow at the same rate of
population since as population grows the set of potential entrepreneurs grows
proportionally (due to a being a parameter). Given Proposition 2-2 and equation 2-16 the
balanced growth rate of innovation is (D = n Using this, c.(t)/c(t) = 0 ,the wage as
numeraire, and both the resource (2-15) and R&D (2-12) conditions, the balanced growth
values of 2 and c^ may be solved for explicitly. The results are graphically represented
in Figure 2-1. Only the positive quadrant is considered since per capital consumption and
R&D difficulty only have values greater than or equal to zero. The R&D constraint is
upward sloping because increased R&D increases quality of goods. The increase in
quality is translated to increased per capital consumption due to decreases in quality
adjusted prices. The vertical intercept of the resource condition is ii. As the per capital
R&D difficulty increases, this signals an increase in required assets needed to innovate to
maintain the same level of industry innovation. As a result, labor resources are shifted
away from manufacturing jobs. With fewer products manufactured, per capital
consumption must decrease, implying a downward sloping resource condition. The two
lines intersect at a unique point identifying equilibrium values, 2 and c .
Ae al-l -1 p
x = (2-17)
(f + ea)Gn(1 p + p[A 1D+ ~pp
c = (2-18)
(n(1- pu + p[il -1D+ pp)
2.4.1 Transitional Dynamics
Since this is a dynamic model, it must be shown that over time the economy can
converge to the equilibrium values stated above when out of equilibrium. To formulate
the first differential equation recall that x(t) = X(t) N(t). Using this and Assumption 2-1,
x(t)/x (t) = pe(t) n is obtained. The industry innovation probability is solved by using
the resource equilibrium condition (2-15). The resulting differential equation for per
capital R&D difficulty is
Aepu c(t) 2-9
+e 1 i
In balanced growth it is assumed that x = 0. Transforming the above equation by
solving for c(t) yields one that is identical to the equilibrium resource condition. It has
already been shown that it is downward sloping with a vertical intercept of Ai.
The per capital consumption differential equation is derived using the maximization
of consumer utility, (2-4). The riskless rate of return is substituted by using the R&D
equilibrium condition. The result of this is
Aec(t) p ct
c(t) = -~>(~ p ct -C 1- c(1- p) pct) (2-20)
By following the balanced growth assumptions the above equation is found to be
upward sloping with a vertical intercept is A(1- pu)/(A pu). This is strictly below the
vertical intercept of (2-19) since Ai > 1 and pue (0,1]. Therefore the two equations
intersect at point E in Figure 2-2.
Increases in x(t) affect (2-19) positively so positive changes will lead to larger x
and negative changes will lead to a reduced x -. This affect is identified by the horizontal
arrows in Figure 2-2. Likewise, changes in c(t) will have an effect on (2-20). In this case
increases in c(t) will result in a decrease of c with the opposite being true for decreases
in c(t). Figure 2-2 shows this effect with the vertical path arrows. As the figure shows,
there exists a saddlepath where the model will transition from out of equilibrium to the
balanced-growth values as described in (2-17) and (2-18).
2.4.2 Economic Growth
The final term of balanced growth to be concerned with is the overall rate of
growth in the economy. This is defined as the rate of growth in consumer utility which is
calculated by using the log of the subutilty function. Substituting in consumer demands
for the highest quality product leaves
logut) =log loga -jb m, tde (2- 21)
The last term of the above equation represents the sum of all quality levels across
all industries multiplied by log ii. The sum of quality levels is analogous to the sum of
innovations that have occurred to date, m(t) = jmr)dv Summing this term across the
number of industries results in the total number of innovations in the economy. Finally,
remember that balanced growth implies that c(t) is constant over time. Differentiating (2-
21) with respect to time yields the growth rate of consumer utility, g. Sub stituting
O(t) = n/pu into the equation to get the balanced equilibrium growth rate of the economy,
g u nlg A (2-2)
While it appears that this is the same balanced growth equilibrium growth rate as in
Segerstrom (1998), there are three main differences. First the productivity parameter pu
has been constrained to be less than or equal to one. Second, the growth rate n, refers to
the growth in entrepreneurs, not the overall population. Finally, the elimination of the
financial intermediary from the economy will result in no innovation. Without
innovation, it is impossible for the economy to grow which highlights the pivotal role of
the financial sector in long term growth.
2.4.3 Comparative Statics
Changes in the parameters will have different affects on the equilibrium values of x
and c. A summary of all the first order conditions for the equilibrium values of a~ and C
appears in Table 2-1. This section focuses on some of the more important results of the
Proposition 2-3: Increases in the probability of being a successful entrepreneur
will increase the amount of per capital innovation.
Changes in the probability a can be a result of advanced educational attainment or
worker training. A better trained work force will increase the likelihood that any
potential entrepreneur will have the skills necessary to innovate. While the model does
not include R&D subsidies like other Schumpetarian models, other government actions
such as student loans, government grants, and increased funding to higher education will
induce a higher a and thus higher growth. Likewise more flexible standards on
evaluating a potential entrepreneur will result in more innovation. The clarification of the
"prudent man" clause can be seen as a loosening of regulations which then allowed more
potential proj ects to be viewed as good investments (see Kortum and Lerner (2000)).
Proposition 2-4: Increases in required financial intermediary employment will
decrease per capital R&D difficulty while increased use of researchers will increase per
capital R&D difficulty. Neither researcher nor financial intermediary employment levels
affect per capital consumption.
An increase in the number of financial intermediary employees signals an increased
cost of evaluation of each entrepreneurial proj ect. The number of positively rated
proj ects will decrease since a higher threshold of earnings is required to offset the
increased rating cost. Since research labor increases the innovation probability it is clear
that an increase in the number of researchers will lead to an increase in innovation and
therefore R&D difficulty, ceterus paribus. Consumers allocate their per capital
consumption independently of prices and quality. Therefore any changes in parameters
that solely affect production will have no effect on per capital consumption. A similar
argument can be made for increased costs.
Corollary 2-1: Increased costs of evaluating an entrepreneur will reduce the
amount of capital provided by investors.
From the model, it is clear that entrepreneurial proj ects require startup capital.
Fewer acceptable entrepreneurial projects leads to less startup capital provided. In
addition, increased investor skepticism may result in increased costs of evaluation. The
recent accounting scandals and dot-com "shake-out" can be put forth as examples that
would increase investor skepticism. These events also corresponded with decreased
levels of investment in new proj ects. Of course, more research must be done to firmly
establish a causal relationship.
Proposition 2-5: Positive productivity shocks through the parameters C1 and A will
positively impact R&D difficulty.
The affect pu has on R&D difficulty comes directly from Assumption 2-1. Further,
increases in A will increase the probability of innovation. As innovation becomes faster,
the R&D difficulty increases. Illustrations of these can be found through the Internet and
increased diffusion of computers. The transmission of information across the Internet has
increased the productivity by decreased time lags and costs. The automation of various
processes through computers is reflected as changes the parameter pu.
2.5 Conclusions and Extensions
This chapter develops a general equilibrium model to explain the relationship
between the finance sector and economic growth. It makes several improvements on the
previous literature. First, the King and Levine (1993b) model of financial intermediaries
has been updated to adjust for population scale effects. Second, the model uses product
rather than process innovations. Third, unlike Segerstrom (1998), not all R&D labor
makes direct contributions to innovation since some must be allocated to the financial
intermediary. As a result of these changes, the model is able to evaluate the impact of the
financial intermediary on economic growth without relying on the level of population--
something that has been empirically shown to lead to inaccurate growth rate predictions.
By removing the scale effects property, a barrier inhibiting the understanding of the
finance-growth relationship is likewise removed.
To highlight the model's intuitive appeal it has been shown that steady-state growth
is affected by changes in the growth rate of entrepreneurs--a direct consequence of how
the finance sector is included in the model. The significance of the finance sector on
economic growth is highlighted in the way all innovations are funded. Without funding
from an intermediary, there could be no innovation. The model also underscores the
importance of education and other factors since they increase potential entrepreneurial
success and thus, innovation. Through parameter shifts, the model is also able to explain
several recent events. Increases in investor doubt, represented by increases in evaluation
cost, will lead to a decrease in the level of investment. Increases in monitoring (or the
effectiveness of monitoring) increase the amount of innovation taking place in the
economy. In each case the predicted outcome matches the actual outcome as experienced
in the United States during the early part of the decade. Accounting scandals were
followed by a decrease in investment and economic growth. Increased concern by
financial intermediaries over their investments excesses led to increased success.
In addition to the above results, the model provides the foundation for several
extensions that would increase awareness of the contribution a finance sector makes to
economic growth. Currently the model assumes perfect information by the intermediary
after the period of evaluation. Incorporating asymmetric information would prove to be
valuable addition to this line of research. The contribution of venture capital to economic
growth has been considered in recent papers by Kortum and Lerner (2000) and Hellman
and Puri (2000). Although King and Levine (1993b) cite the venture capital process as a
motivation for their model, to directly translate the results of their model and this one as
the impact of venture capital would be an exaggeration of the impact of venture capital
investments. Adding multiple financial intermediaries to the existing framework would
advance the understanding of venture capital's role in economic growth.
Figure 2-1. Equilibrium Conditions for Model
Figure 2-2. Stability of Balanced-Growth Equilibrium
Table 2-1. Comparative Statics
Comparative statics are the partial derivative of the equilibrium values (Equations 2-17
and 2-18) with respect to the parameters stated above. By assumption, pue (0,1]. If C1=1,
then 89/8p and 88/8p both equal zero but all other effects are the same.
THE EFFECT OF PRUDENT INVESTOR LAWS ON INNOVATION
As early as 1985, but predominately in the mid-1990s, states began to adopt new
laws that govern investments made by Eiduciaries. The laws were direct enactments of
the Uniform Prudent Investor Act (University of Pennsylvania) as drafted by the National
Conference of Commissioners on Uniform State Laws or only marginally different. The
primary goals of the laws were to introduce modern portfolio theory as it applies to the
prudence of investments, remove restrictions on specific Einancial instruments, and to
allow delegation of management. Previous empirical investigations on related subjects
have shown that laws regulating Eiduciaries affect their portfolio holdings, that previous
reductions of prudence standards for pensions led to increased investment in venture
capital, and that the Einance sector can positively affect technological change and
economic growth. Many theoretical models have also shown the positive affect of
Financial intermediaries on technological change. These previous theoretical and
empirical findings suggest that the changes in prudent investor laws should alter the
composition of investments and lead to increased technological progress. Few papers
have considered the impact of prudent man or prudent investor laws. Of those that do,
none consider the impact of prudent investor laws on something other than equities. This
chapter considers the impact of the law on innovation by considering its impact on
venture capital, R&D, and patenting and therefore provides a significant expansion to the
A variety of regression specifications and models are used to determine empirically
the effect of the new prudence regime. The basic specification considers how R&D,
venture capital investment, and patenting are affected after a state adopts the law. These
regressions ignore the complexity of spillovers across states and how timing of adoption
affects the outcomes. Other regressions included variables that identify neighboring
states that had adopted prudent investor laws in order to account for any interstate
spillovers. Regressions were run on state quartiles since the effect of the law change
should be different for states with different shares of venture capital, R&D, or patenting
or for those states that adopted the law later. The maj ority of evidence shows venture
capital, R&D expenditures, and innovation (as proxied by citation weighted patent
counts) were unaffected by the new prudence standards. In regressions for states in the
bottom quartiles of R&D and patenting there is a slight positive effect, but nothing is
found for venture capital. While previous research has shown changes to the prudence
standards of pension funds had a significant impact on innovation, given the small size of
the impact, the same cannot be said for funds held it trust by banks. The results presented
here provide a significant contribution to the literature on the finance-growth nexus by
showing the effectiveness of financial intermediaries in fostering innovation is bounded
and that not all policy reforms in the United States favorable to financial intermediaries
are equally capable of fostering innovation.
In section two of this chapter, the background of prudent investor laws are
reviewed to underscore the consequences of adopting the new prudence standards. It
continues by summarizing relevant literature studying prudent man, prudent investor, and
the relationship between finance and technological change. The third section reviews the
data sources and empirical methodology. Section four presents benchmark analyses to
provide the intuitive foundation for the main empirical results. It also considers tests the
potential endogeneity of the prudent investor law. Section five presents the main
empirical results of the impact on venture capital, R&D expenditures, and patenting using
detailed data and more advanced techniques. The final section offers conclusions and
implications of the findings.
3.2.1 Prudence of Investmentl
Laws governing the investments offiduciaries have long been based on the abstract
notion of prudence. The origins of this yardstick date back to 1830 in Harvard College v.
Amory. The opinion from the Massachusetts Supreme Judicial Court for that case stated
"[a trustee] is to observe how men of prudence, discretion and intelligence manage their
own affairs" and then act similarly (Longstreth (1986, p. 12)). In spite of being around
for nearly a century, the prudent man rule was not the rule of law for a maj ority of states
until the mid-1940's. Prior to that, many states chose a more strict approach by
specifying lists of acceptable investments for trusts.2
In the middle of the 20th century, the Model Prudent Investment Act (MPIA),
backed by the American Banking Association was adopted by many states. The MPIA
contained language, nearly verbatim, from the original Harvard College v. Amory ruling
and made 'prudence' the standard in evaluating investments. Rather than specifying a list
of investments that were acceptable, the prudence of an investment was to be decided on
SThis sub-section draws from Longstreth (1986), Shattuck (1951), and Langbein (1995).
2 See Shattuck (1951, p. 502-504) for a detailed classification of state statutes pre- and post-1940.
a case by case basis. The MPIA did, however, explicitly forbid the use of investments
such as those in "new and untried enterprise" which were deemed speculative and
therefore imprudent (Langbein 1995). In spite of these apparently more liberal rules, the
interpretation by courts and subsequent cases of precedence led to rules just as restrictive
as legal lists (see Restatement of the Law (1992)).
The relative impotence of some fiduciaries caused by prudent man laws and their
interpretation led to the creation of the Uniform Prudent Investor Act (UPIA). The law
made five substantive changes to the evaluation of"prudence" and management of
investments. One of the more significant changes was the inclusion of modern portfolio
theory in gauging prudence:
A trustee's investment and management decisions respecting individual assets must
be evaluated not in isolation but in the context of the trust portfolio as a whole and
as a part of an overall investment strategy having risk and return obj ectives
reasonably suited to the trust. (University of Pennsylvania @ 2(b))
The UPIA further stated that no type of investment can be categorically deemed as
If the predictions hold true, the adoption of prudent investor laws will have a large
impact on venture capital investment, R&D expenditures, and innovation. According to
Flow of Funds data from the Federal Reserve,3 COmmercial banks in the U. S. held an
average of $1,358.64 billion in treasury securities, municipal securities, corporate
equities, or mutual fund shares between 1995 and 2004. Much of the investment was
made using money held in trust. It is unlikely that the adoption of prudent investor laws
will cause a complete shift away from these asset types to just venture capital or new and
untried enterprises. However, even a shift of 5% would result in an additional $67 billion
3 Board of Governors of the Federal Reserve System (p. 61), line 5 minus line 10.
investment in R&D-about $12 billion more than was spent on R&D in California in
2000. While banks are directly under the jurisdiction of UPIA it is possible that the law' s
impact is broader. Survey evidence from Longstreth (1986) suggests that pension fund
managers were also constrained in their investment choices by prudent man regulations.
This is in spite of the fact that pension funds have had more liberal prudence guidelines in
place since 1979. One explanation for this is that case law and precedents based on
Restatement of the Law (1959) were common while the same could not be said for more
recent laws governing other fiduciaries. Trustees were thus limiting their investments to
some degree in order to protect themselves from uncertain litigation outcomes
(Longstreth, 1986). With prudent investor guidelines, this uncertainty may fade and
other fiduciaries may use a different investment strategy.
3.2.2 Previous Literature
Previous studies by Del Guercio (1996) and Gompers and Metrick (2001) explore
the impact of the restrictive prudent man guidelines. Del Guercio examines the impact
prudent man laws have on the management of equities by institutional investors. She
observes different reactions by bank and mutual fund managers. While bank managers
are more likely to shift their portfolios toward stocks "viewed by the courts as prudent,"
mutual fund managers do not. It concludes that by forcing intermediaries to protect their
liability, these laws alter incentives and lead managers to act in ways contrary to the best
interest of clients. Gompers and Metrick consider prudent man's effect on institutional
investor behavior and equity prices. They find some evidence that prudent man
regulations affect the ownership of stocks in favor of older and larger firms. In other
words, those investments that are less risky or more prudent. More recently, Hankins,
Flannery, and Nimalendran (2005) investigate the impact of lessening the prudence
standard on equity holdings. They find that the adoption of prudent investor laws affects
institutional portfolios in the predicted direction. Since prudent investor laws weaken the
duty of fiduciaries, there is a shifting away from dividend paying stocks to more risky
assets. This finding is a logical corollary to Del Guercio's and Gompers and Metrick's
analyses. Gompers and Lerner (1998) investigate the changes in the levels of
investments made by venture capital firms from 1972 to 1994. Their approach considers
fluctuations in investment levels as a result of shifts in the supply and demand for venture
capital. One of the more significant shifts in the supply of venture capital resulted from
the 1979 ERISA clarification which allowed pension funds to invest in venture capital.
The Department of Labor' s clarification and the UPIA redefine prudence similarly so the
enactment of prudent investor laws may also be considered a positive shift in the supply
of venture capital. Finally, Kortum and Lerner (2000) investigate the contribution of
venture capital investment to innovation. They find that venture capital dollars are about
three times more likely to lead to innovations than other forms of R&D financing. Thus,
even a small shift in the supply of venture capital investment resulting from UPIA should
significantly increase innovation. The estimate is based on the results of reduced form
and structural regressions, as well as regressions that use the 1979 ERISA clarification as
an instrumental variable to account for the potential endogeneity of venture capital
The question posed by Kortum and Lerner (2000) is slightly different than the one
posed here. Rather than considering the role of venture capital dollars, this essay
considers the impact of a law that should have affected the type of investments banks
could make with funds held in trust. Thus, this paper tests the impact of the law change
on venture capital, R&D, and innovation rather than testing the contribution of the entire
venture capital industry on innovation. The different tack also implies that the same
concern over the endogeneity of venture capital found in Kortum and Lerner do not
necessarily apply here. In spite these differences, the theoretical model developed in
their paper can be used to illustrate how prudent investor laws should increase the amount
of venture capital as well as the total innovative effort.
Kortum and Lerner (2000) assume that there are costs associated with venture
capital funds (i.e. screening opportunities, recruiting managers, etc.) that may affect the
type of proj ects funded by venture capitalists. As these costs are reduced, more and more
proj ects are capable of being funded by the venture capitalist rather than by traditional
corporate R&D. Kortum and Lerner motivate their use of the ERISA clarification by
stating it "lowered the cost of funds to venture capitalists" (p. 685) which, in their model,
was shown to increase both the total innovative effort and the ratio of venture capital to
corporate R&D. Since the ERISA clarification and prudent investor laws are similar, the
adoption of the new prudence standard is predicted to reduce the cost of funds and thus
lead to an increased share of venture capital and more patenting.
3.3 Data and Empirical Methodology
When taking the previous literature as a whole, the predicted effect of prudent
investor laws on innovation is clear. Less stringent guidelines should cause a shift away
from established assets in favor of other investments, including those in new and
innovative companies. The possibility of such a shift was underscored in Restatement of
the Law (1992) which specifically mentions venture capital funds and new and untried
enterprises as newly acceptable investments under the updated prudence regime. It is
through these two mechanisms that the adoption of prudent investor laws should cause
innovation. The aforementioned theories suggest that money will be used for R&D
purposes which will then lead to innovation. To test the impact of prudent investor laws
on innovation it is necessary to consider the adoption's impact on venture capital, R&D,
and patenting. By doing so the empirical analysis may also test the viability of the above
theories. If patenting were affected, but not R&D or venture capital, then an alternative
and unknown mechanism is causing a change in innovation.
Data on R&D expenditures in a state for the years 1993-2002 is taken from a
survey from the National Science Foundation (National Science Foundation). The time-
period used is limited due to inconsistencies in survey methodology and missing data.
Prior to 1993, new sampling was not done for each survey year. For these non-sampling
years, the National Science Foundation interpolated data from the previous sample. After
1993, sampling was done every year the survey was given. This methodological break
causes pre- and post-1993 survey data to be inconsistent. To bypass this problem, only
data from 1993 forward was used. Another cause for concern is that surveys were not
given annually until 1997, so data for 1994 and 1996 are not available. Further, some
state R&D information was not made publicly available due to privacy restrictions. This
is more likely to affect smaller states since there is anonymity in numbers. Instead of
interpolating for missing values (and perhaps introducing errors in the dataset) data was
taken as-is and missing data caused the exclusion of that observation in regressions. The
dollar values are in billions of 2000 dollars. They were deflated by the implicit GDP
deflator as provided by the Bureau of Economic Analysis.
Venture capital data is from the MoneyTreeThl (PriceWaterhouseCoopers, et al.)
survey jointly sponsored by PriceWaterhouseCoopers, Thomson Venture Economics, and
the National Venture Capital Association. The survey collects information on venture
capital investments made within a state for each quarter starting in 1995. For this
analysis, total annual venture capital disbursements in each state for 1995-2001 were
used. The investment amounts were deflated similarly to R&D expenditures and are in
millions of dollars. There is no indication that the venture capital survey suffers from the
same methodological inconsistencies as the R&D survey, however, the original dataset
does not distinguish between zero investment and unknown investment in a state. If the
survey does not report a dollar value for investment for a particular state in a particular
year then that state is excluded from regressions.
Patent statistics are used as a proxy for innovation. In summarizing previous
research on patents, Griliches (1990) describes how patents weighted by citations may be
a better proxy for innovation than raw patent counts. Patent data from the NBER patent
database was updated by Bronwyn Hall (Hall). Her updated dataset contains detailed
information on patents granted from 1963 to 2002. Due to data limitations on application
year and suspect patent information for 2001-2002, only successful patents applied for
from 1990-2000 are considered.4 The citation weighted patent counts were then summed
to get patent counts by state (including Washington, D.C.) and year of application.
Patents were assigned to a state based on the location of the first inventor listed on the
patent application.' Patent applications from foreign individuals were dropped and
4 Patent information for later years in the database is unreliable due to the time lag between applying for
and being granted a patent. Since the database is based on information on granted patents it is likely that
2001 and 2002 patent information is understated as some applications that are patent-worthy have yet to be
awarded a patent.
5 It is common to assign patents to states in this manner. Other approaches to identifying geograpluc origin
exist: most coming from the literature on geographic knowledge spillovers (see Thompson and Fox-Kean
records with no state information or with values of 'US', 'APO', or US territories were
eliminated. Tables 3-1 through 3-3 provide summary statistics for the data used.
Adoption of prudent investor laws are represented by an indicator variable which
takes on the value of one if a prudent investor law is active in that state during that year.
As shown in Table 3-4, the adoption of prudent investor laws occurred, by and large,
after 1995. By the end of 2000, forty states including Washington, D.C. adopted the
UJPIA. It is possible that investment originating in one state may spillover to neighboring
states. To account for this possibility some regressions included a variable representing
the percent of neighboring states that adopted prudent investor laws. It is plausible that
the neighborhood effects are not linear, i.e., over time the spillovers from states may
increase or decrease exponentially. Select regressions include the squared percent of
neighboring states to account for this possibility. Interstate spillovers were also
accounted for by limiting the regressions to a subset of states.
3.3.2 Empirical Methodology
The empirical analysis uses several different specifications. The fourth section of
this chapter presents the results of a hazard model used to explore the determinants for
the timing of adoption. It also presents estimates on the impact of adoption using a long
difference. The hazard model specification (Equation 3-1) considers how patenting,
among other things, affects the adoption of prudent investor laws and can provide a check
on whether the adoption of these laws is endogenous to the innovativeness of a state.
InLAGADOPTi = a(t) +XiP + i (3-1)
The dependent variable is the difference between the year of adoption and 1980
(the start of the sample). The vector of independent variables, X;, includes average
number of patents for each state from 1980-1989, average Americans for Democratic
Action (ADA) scores for U. S. senators from 1979-1997 (ADA), state average median
income from 1984-1989, and average size of the state's finance sector as a percent of
gross state product. More detail on the sources and predicted signs are provided in
section four of the chapter.
Long differences are also estimated and the results are presented in section four.
These regressions are included to provide a basic understanding of how prudent investor
regimes affected innovative inputs and outputs without having to account for yearly
fluctuations. Each regression uses the specification:
1t~,l ,, -1, = ao + a, Yt+s-k,l t-k,l ]+02 [t+s,l Ptol ]+ 2S + er so -6t,I i, (3 -2)
where Y may be R&D, venture capital, or raw patent counts. The lag for venture capital
and patenting was one year (k=1) but since R&D data is available every other year the
lag was two years (k=2). The difference in prudent investor variables may be zero or 1
but never -1 since no state changed back to a prudent man regime. Venture capital
information was not available for all states for the same years. For example, the level of
venture capital investment in Texas was known for 1995 through 2004 but was only
known for North Dakota for 1997 through 2004, therefore the number of years the
difference encompasses may be different by state. A vector of indicator variables, S,
were included to account for these issues. The vector includes indicator variables for
differences of 7 years, 8, years, and 9 years. Since North Dakota data uses a seven year
difference, the indicator variable representing seven years was equal to 1 and all others
were equal to zero. For Texas, the indicator variable representing nine years was equal to
1 and all others were equal to zero. Other states were similarly recorded.
The main analysis in section Hyve uses the specification:
Y,, = oY,,_z + P,PI,,z + L'P + e,,. (3 -3)
As before, the dependent variable Y may be venture capital, R&D expenditures, or
citation weighted patent counts. A vector of variables on the adoption characteristics of
neighboring states is represented by L. These include year indicator variables, the
percentage of neighboring states that adopted prudent investor laws, and that percentage
squared, although not all regressions included all variables. As was true for the long
difference, R&D data requires that the dependent variable is lagged two years while the
other data uses only a one year lag. The regressions use the Blundell-Bond estimator as
implemented in the xtabond2 module for STATA.6 This estimation procedure was
necessary due to the lagged dependent variable on the right-hand side of Equation 3-3
and the potential for time-invariant characteristics The Blundell-Bond estimator (known
elsewhere as GNINBB Or B-B) uses the first difference of the empirical specification to
remove time-invariant characteristics. It then uses the lagged levels of the dependent
variable as an instrumental variable for the lagged difference of the dependent variable.
Since these instruments alone may not be effective, the estimator also uses lagged
differences as instrumental variables for lagged levels. For this research the endogenous
variables are the prudent investor indicator, neighbor state effects, and the lagged
dependent variable. Year effects are also included in both the dynamic panel regressions
and as exogenous instruments. Tables 3-7 through 3-9 report the results for these
regressions. In addition to coefficient estimates, the tables include p-values for the
Arellano-Bond test for autocorrelation and the Sargan/Hansen test for overidentification.
6 MOre information on the xtabond2 module may be found in Roodman (2005). Appendix B provides a
more thorough treatment of the Blundell-Bond estimator and the reasons why it was preferred over OLS.
For the Blundell-Bond estimates to be consistent, second order serial correlation must not
exist (Arellano-Bond statistic rej ects the null) and the model must not be overidentified
(Sargan/Hansen test statistic must rej ect the null of overidentification). Regressions are
one-step GMM unless otherwise noted.
The sample was divided into quartiles and regressions were run separately for
quartiles of venture capital, R&D, and citation-weighted patent counts. By dividing the
sample into quartiles of the respective dependent variable, it is possible to see how the
law affected states with different sized venture capital, R&D, or innovative output. Only
the top quartile and bottom quartile are presented. The sample was also broken into
quartiles based on the adoption of the law. It is likely that the first states to adopt the law
should be those where the impact, if any, is the greatest. Regressions for the bottom
quartile of adopters were also included to estimate the potential differences prudent
investor laws had across states.
3.4 Tests of Exogeneity & Benchmark Regressions
3.4.1 State Innovative Output and the Timing of Adoption
The UPIA was available for adoption by all states at the same time yet the adoption
of the law was not uniform across states. That some states adopted the law earlier than
others may be exploited as a means of understanding who considers the law important. It
also allows a check of whether or not highly patenting states adopted the law first. If they
did, then there is an endogeneity problem. The specification in Equation 3-1 was used
for all survival analyses.
Estimating survival functions was complicated due to right-censoring and ties.
Data was right censored after 2000 since data on patents is incomplete after that year.
Several ties occurred in the data since several states also adopted the law in the same year
which may affect estimation. Proportional hazards models were used for all estimates to
account for these problems. Due to the small sample size and low computational cost,
exact estimation methods for ties were used (for more information see Allison (1995, p.
127)). The a(t) term is eliminated by using the proportional hazards model.
Americans for Democratic Action scores were used as a proxy for a state's political
leanings. They were taken from Francis and Kenny (1999, pp. 88-89). If a state had a
senator from both parties, the average of the two scores was taken. For states that only
elected members of one party, the ADA score used was the score of that party. The
predicted sign of the ADA coefficient is not obvious, ex ante. On one hand, more liberal
states may feel that prudent investor adoption may only benefit the wealthy. Therefore a
higher ADA score will lead to a longer lag in adoption. On the other hand, more liberal
states are also those with higher levels of income so they may be more likely to adopt
prudent investor laws. To make this relationship more clear, median income is also
included in some of the regressions. Finally, if the finance sector is a large part of gross
state product it is more important to adopt favorable legislation and do so early.
Therefore the coefficient for this variable is expected to be positive.
Averages of patents across the 1980's were used for several reasons. With the
exception of Delaware, no state adopted prudent investor laws prior to 1992. Therefore,
including information from the 1990's may lead to false precision since adoption of the
laws may have affected rates of innovation.7 Adoption of the law is unlikely to be a
reaction to one year' s worth of patent statistics. If patenting did play a role in the
decision to adopt, it is more likely that historical patent rates were considered. Finally,
SAn indicator variable for Delaware was also considered due to its unique nature. It was not, however,
checks of the data confirmed that despite fluctuations in the amount of patenting, the
ranking of states patenting outcomes did not change much. Since survival analysis uses
rank order, the average is an acceptable method of approximation.
Table 3-6 shows the results of the regressions. Regardless of specification, the
predictive power of a state's average patenting on adoption is indistinguishable from
zero. The coefficient with the most explanatory power belongs to the relative size of the
Einance sector in a state, which is not surprising. An industry with large economic
significance is likely to have the political capital necessary to influence legislative action.
The ADA score coefficient is statistically significant when median income or the size of
the finance sector is excluded. Like average patents, median income is not statistically
significant in any regression. That patenting is not significant suggests that the potential
innovation effects were not a consideration in adopting the law. Thus, the prudent
investor law is assumed to be exogenous in the regressions presented in section five.
3.4.2 Evidence from a Long Difference
Before considering the affect on annual data, it is instructive to see if the change of
the dependent variable over a long period is influenced by the prudent investor law. If
the regressions on the long difference indicate a statistically significant impact then the
impact of the law change seen in more detailed regressions should likewise be large.
However, if the coefficient estimates are not significant, then any impact of the prudence
regime change should be small or nonexistent. The regression results are presented in
The change of prudence regimes appears to have no impact on the change in
venture capital, research expenditures, or patenting. Based on the R-squared, the venture
capital and patenting regressions appear to fit the data well. The regression on R&D
expenditures is not as well estimated, this is undoubtedly a result of the every-other-year
nature of the data. These basic regressions are the first (of much) evidence to suggest
that prudent investor laws have not had the predicted affect on innovation.
3.5 Prudent Investor Laws & Innovation
3.5.1 Indirect Investments: Venture Capital
The most obvious mechanism by which prudent investor laws should affect
innovation is through venture capital. Banks invest in venture capital funds which in turn
invest in innovative companies. Therefore changes in investments by banks will have an
indirect affect on innovation. Investment in venture capital causes several difficulties for
the analysis. Unlike R&D, there is no intuitive or legal reason that money originating in
one state must stay within that state's borders. In fact, Florida and Smith (1993) Einds
that venture capital investment gravitates towards regions with an established venture
capital sector. If this is the case, then Minnesota' s adoption of a prudent investor law, for
example, may not impact the amount of venture capital invested in that state. Instead
there may be a flow of money to California. Fortunately, Florida and Smith's findings
also suggest a method on how to deal with the problem of interstate flows. Just as
investments in a state with a small venture capital sector will flow outside of the borders,
the strong pull of a large venture capital sector in a state should keep investments inside
its borders. That is, Minnesota may have trouble keeping its venture capital investment
within its borders but California likely does not. A state with a small venture capital
sector may likewise Eind a negative or zero effect of the prudent investor law.
To consider this issue, states were divided into quartiles based on the size of their
venture capital sectors. The first quartile (states with large venture capital sector) and the
fourth quartile (states with the smallest venture capital sector) were considered
separately. The results are presented in Table 3-8. As predicted, the bottom states
(fourth quartile) observed no impact. Surprisingly however, neither did the top venture
capital states. Another way of dividing the sample is to look at when they adopted the
prudent investor law. That is, whether the state was among the earliest or among the
latest adopters of the new prudence regime. It was not possible to test the impact on the
early adopters, since the data on venture capital begins several years after they adopted
the law implying no variation in the prudent investor indicator variable. It was possible
to consider the late adopters. As predicted (regressions 7 through 9 of Table 3-8), the
coefficient on the prudent investor indicator has a negative sign. Without complementary
evidence from the set of early adopting states it is difficult to interpret the results on the
late adopters. It might be that, as late adopters, these states were crowded out of
collecting venture capital investment. This may have less to do with the prudent investor
law than to do with the timing of adoption; a causal versus correlation issue.
Confirmation from regressions on R&D or citation weighted patent counts are necessary
before drawing any firm conclusions.
3.5.2 Direct Investments: R&D Expenditures
In addition to venture capital investment, the UPIA allows direct investment by
banks to new and untried enterprises. This would represent a direct investment of funds
held in trust by banks that may encourage innovation. In this regard, banks are likely to
act as a venture capitalist would. A large body of research on venture capital (Lerner
(1995), Sorenson & Stuart (2001), and Zook (2002), among others) shows that venture
capital investment is local. That is, venture capital funds invest in firms that are close to
the offices of the venture capitalist in order to provide better guidance to the (perhaps)
uninitiated entrepreneur and protect their investment. If banks are directly investing in
new and untried enterprises then it is logical to assume they follow the same strategy for
similar reasons, therefore the spillover problem seen with venture capital likely does not
affect direct investment by banks.
Of course "local" does not imply "intrastate" since many investments may cross a
state's borders, although, they will be confined to neighboring states. Most regressions
include the percentage of neighboring states that have adopted prudent investor laws to
capture these effects. Percentage terms are needed since not all states are bordered by the
same number of states. Inclusion of the neighbor state impacts causes the exclusion of
Hawaii and Alaska from regressions.
Estimates of the prudent investor law' s impact on direct investment may be found
in Table 3-9. Ignoring regressions (1), (4), and (10) due to the low p-value of the Hansen
statistic, the measured impact of the prudent investor law on R&D is mixed. As with
venture capital, the larger level impact should be found in the Top R&D states, states
with the highest amount of per capital expenditures. In percentage terms, those states with
the least amount ofR&D expenditures should see gains. The latter is confirmed in
regression (6) of Table 3-9. When accounting for interstate spillovers, there is a positive
and statistically significant effect of prudent investor laws for the bottom R&D states.
However, their appears to be no impact of the prudent investor law on the top R&D
As before, the data was divided into quartiles based on when each state adopted the
UJPIA. The evidence for the early adopters is ambiguous since, each specification leads
to a different result: positive, negative, or zero. For late the estimates are based on the
third quartile since, for the fourth quartile, the prudent investor indicator remains
constant. Again, ignoring regression (10) due to the poor Hansen statistic, there appears
to be no impact of the prudent investor law on the R&D of late adopters.
3.5.3 Alternative Mechanisms
The theories cited in section two indicate that prudent investor laws will impact
innovation through the mechanisms of venture capital and R&D. The preceding analysis
tested these mechanisms for any change that could be associated with the adoption of
prudent investor laws and found some weak supporting evidence, but only for the late
adopters of the law for venture capital or those states with the lowest R&D output.
Neither of these findings shows a strong positive impact on innovation that previous
theoretical and empirical work predict. This final section considers the impact on
innovation through any mechanism, not just venture capital or R&D.
Citation weighted patent counts for each state are used to estimate the impact of the
prudent investor law on innovation through any means. As before, states were analyzed,
by quartiles, based on the amount of innovation that takes place within that state' s
borders and the timing of adoption. The analysis for late adopters is based on the third
quartile due to no variation in the prudent investor indicator for the fourth quartile. For
all but one regression, the results indicate the prudent investor laws had no effect on
innovation. The sole exception, regression (5) of Table 3-10, shows that the prudent
investor law had a small, positive impact on innovation for the least patenting states.
This is treated as confirmation of the R&D results in the previous section. If the bottom
R&D states were positively affected, then the likely consequence of this is that patenting
in bottom states would also increase. This is the only evidence that suggests that prudent
investor laws had any effect on innovation. That the impact is weak and not in the states
expected suggests that previous results considering the impact of venture capital on
innovation are likely overstated.
Prior empirical and theoretical research suggests that the adoption of prudent
investor laws should have an unambiguously positive effect on innovation. The evidence
presented here, by and large, shows otherwise. While some regressions suggest a small
positive impact, most estimates indicate that prudent investor laws have had no impact on
venture capital, R&D expenditures, or innovation. The latter results can be interpreted
either optimistically or pessimistically. The optimistic scenario is that fiduciaries have
yet to fully extend their investments to the point they are legally allowed. This view
suggests that states that have adopted prudent investor laws should experience
technological change and economic growth sometime in the future. The pessimistic view
interprets the results as evidence that prudent investor laws do not affect technological
change. That is, prudence guidelines for Eiduciaries were already sufficient to promote
growth before prudent investor laws were enacted.
It is likely that the latter scenario is the correct description of events. Under the
optimistic view, investment is constrained while case law is established. If this were true,
investments, including those in equities, would not be impacted until well after the
adoption of these laws. This is contradicted by Hankins et al. (2005) who show equity
investments have already been affected by prudent investor laws. The pessimistic view is
consistent with previous empirical findings on prudent investor laws but runs counter to
prior theoretical and empirical work. The results presented here suggest that not all laws
affecting investment decisions will lead to changes in technological change, for good or
ill. Further, not all Eiduciaries will be able to positively impact innovation.
This research, as with most, does not close the door on further investigations on the
topic. Estimates on the impact of prudent investor laws would greatly benefit from
micro-level data on financial intermediary investments. The detailed data could expand
the analysis to include systems of equations rather than changes in the equilibria.
Due to variation in the time series available for data, correlations are based on 1990-2000 values.
Table 3-1. Correlation Matrix
Citation Weighted Patent Counts (WPC)
WPC, lagged 1 year
WPC, lagged 2 years
Log of deflated VC disburmsents (VC)
VC, lagged one year
VC, lagged two years
Log of deflated R&D expenditures (R&D)
R&D, lagged one year
R&D, lagged two years
Prudent Investor Indicator (PI)
PI, lagged 1 year
PI, lagged 2 years
WPCt WPCtz WPCt-2 VCt
VCtz VCt-2 R&Dt R&Dtz R&Dt-2 PIt PItz PIt-2
0.0876 0.1018 1.0000
-0.0337 -0.0184 0.7671 1.0000
-0.0584 -0.0656 0.6852 0.8932 1.0000
Table 3-2. Descriptive Statistics
Variable Obs Mean Dev. Min Max
Prudent Investor Indicator 561 0.299 0.458 0 1
Log, Deflated VC disbursments 280 4.427 2.304 -3.080 10.667
Log, Deflated R&D expenditures 343 -0.147 1.919 -6.178 3.8236
Citation Weighted Patent Counts 561 6395.18 13090 3 106580
Descriptive statistics are based on 1990-2000 data.
Table 3-3. Descriptive Statistics by Year
Citation Weighted Patent Counts
Lon, Deflated VC disbursments
Log, Deflated R&D Expenditures
Mean Median Deviation
Mean Median Deviation
Year Mean Median Deviation P
1990 10,047 4,283 15,798
1991 9,734 4,111 15,600
1992 9,809 4,098 16,496
1993 9,281 3,760 16,362
1994 9,044 3,371 16,122
1995 8,425 3,170 15,492
1996 6,223 2,298 11,954
1997 4,555 1,536 8,713
1998 2,152 786 4,027
1999 890 308 1,648
2000 185 72 324
Descriptive statistics are based on 1990-2000 data.
Table 3-4. Year of Adoption of UPIA (or equivalent)
State Adoption State
Delaware 1985 Alaska
Illinois 1992 Vermont
Florida 1993 Washington, D.C.
Maryland 1994 Massachusetts
New York 1995 New Hampshire
South Dakota 1995 Ohio
Colorado 1995 Indiana
New Mexico 1995 Wyoming
Oregon 1995 Pennsylvania
Utah 1995 North Carolina
Washington 1995 Virginia
Oklahoma 1995 Michigan
California 1996 Iowa
Arizona 1996 Kansas
Montana 1996 South Carolina
Rhode Island 1996 Tennessee
West Virginia 1996 Nevada
Nebraska 1997 Texas
New Jersey 1997
North Dakota 1997
Source: Hankins et al. (2005)
Table 3-5. Comparison of Adopters and Non Adopters
Avg. Size of
Table 3-5. Continued
Avg. Med. Avg. Industrial
Adoption Household Avg. Avg. Size of R&D
Year Income Population Avg. GSP Finance Sector Expenditures
1985 $23,177 4,746,110 $82,836 $14,034 $2,585
1992 30,463 4,953,150 119,334 21,540 2,520
1993 31,147 4,825,907 121,054 21,814 2,487
1994 32,213 4,869,498 128,572 22,611 2,517
1995 33,761 4,945,554 134,834 22,606 2,569
1996 35,551 4,551,691 129,586 20,419 2,039
1997 36,419 5,264,710 157,522 9,297 2,125
1998 37,603 5,719,932 178,305 10,966 2,298
1999 38,822 6,300,352 200,875 11,805 2,208
2000 38,620 6, 156,757 195,849 10,842 1,314
2001 38,839 6,447,856 211,176 12,139 1,371
2002 39,553 6,607,460 221,133 13,073 1,412
2003 39,021 7,246,438 251,296 15,300 1,548
2004 .5,169,695 180,490 10,648 882
Industrial R&D expenditures is for 1995 data only since annual data is not always available. They are reported in thousands of current
dollars. GSP and Finance sector data pre- and post-1997 are inconsistent due to switch to NAICS classifications. GSP & Size of
Finance Sector are in millions of current dollars. Median Income is in current dollars.
Table 3-6. The Timing of Adoption
ADA score "
Avg. Median Income
Finance Sector as a
Percent of GSP
Wald 1.179 6.266 8.2254 14.6295 14.4011 14.6167 14.3869
LAGADOPT is dependent variable. Standard errors are reported in parentheses.
a Inclusion of ADA scores decreases the number of observations by 1 since the District of Columbia does not have Congressmen.
Table 3-7. Impact of Prudent Investor Laws over Long Difference
Lagged Difference Yr+sl-rft-,
Lagged Prudent Investor Difference PI,,s_1-PI,;
Indicator Variables: Sizes of s
Ob servati ons 50 51 51
R-squared 0.98 0.71 0.98
The dependent variable is Yt+s-Yt, where Y is Venture Capital Investments, R&D Expenditures, or Raw Patent Counts for each state.
R&D expenditures and Raw Patent Counts were available for all states for the same period of time. Sizes of s were included in the
V.C. regression due to differences in the availability of data across states.Robust standard errors in brackets. p<0. 10, ** p<0.05, ***
Table 3-8. Estimates of the Impact on Venture Capital Investments in a State
Top VC States (1st Quartile) Bottom VC States (4th Quartile) Late Adopters (4th Quartile)
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Log of deflated VC investment (lagged one year) 0.9664*** 0.9695*** 0.9693*** 0.4118*** 0.5411*** 0.5414*** 0.8797*** 0.9362*** 0.9357***
[0.0263] [0.0249] [0.0264] [0.1399] [0.1412] [0.1418] [0.0662] [0.0522] [0.0543]
Prudent Investor Indicator (lagged one year) -0.023 -0.0393 -0.0389 0.3889 0.1548 0.1558 -0.4352 -0.4561* -0.4563*
[0.0364] [0.0261] [0.0325] [0.5042] [0.4768] [0.4729] [0.2747] [0.2534] [0.2594]
Percent of neighboring states adopted (lagged one year) 0.0151 0.0203 0.0895 -0.2875 0.1983 0.3048
[0.0555] [0.1913] [0.5840] [2.0950] [0.1322] [0.6160]
Percent of neighboring states adopted, squared (lagged one year) -0.0053 0.3192 -0.1293
[0.1767] [1.9806] [0.6409]
Year Indicator Variables Y Y Y Y Y Y Y Y Y
Observations 72 72 72 48 42 42 60 60 60
Number of states 12 12 12 12 10 10 10 10 10
Hansen statistic (P-value) 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000
Arellano-Bond, AR(1) (P-value) 0.049 0.049 0.050 0.359 0.222 0.221 0.076 0.084 0.084
Arellano-Bond, AR(2) (P-value) 0.240 0.233 0.236 0.383 0.107 0.102 0.809 0.835 0.837
* p<0.10, ** p<0.05, *** p<0.01. Robust standard errors are in brackets. Regressions were run using the Blundell-Bond dynamic
panel estimator. The symbol 'Jf' indicates a two-step robust GMM estimation. All other regressions are robust one-step regressions.
Alaska and Hawaii are dropped when including data on neighboring states. Top VC States (1st Quartile): Illinois, Florida, New York,
Colorado, Washington, California, New Jersey, Massachusetts, Pennsylvania, Virginia, and Texas. Bottom VC States (4th Quartile):
South Dakota, New Mexico, Montana, West Virginia, Arkansas, Hawaii, Idaho, North Dakota, Alaska, Vermont, Wyoming, Iowa, and
Nevada. Late Adopters (4th Quartile): Delaware, Illinois, Florida, Maryland, New York, South Dakota, Colorado, New Mexico,
Oregon, Utah, Washington, and Oklahoma. Late Adopters (4th Quartile): Pennsylvania, North Carolina, Virginia, Michigan, Iowa,
Kansas, South Carolina, Tennessee, Nevada, and Texas.
Table 3-9. Estimates of the Impact on R&D Expenditures in a State
Top R&D States (1st Quartile) Bottom R&D States (4th Quartile)
(1) (2) (3)t (4)t (5)t (6)t
Log of deflated R&D expenditures (lagged two years) 1.1801*** 0.9248*** 1.0988*** 0.7670*** 0.6148*** 0.8146***
[0.1404] [0.0942] [0.1056] [0.1878] [0.1820] [0.2332]
Prudent Investor Indicator (lagged one year) -0.414 0.0597 -0.1708 1.1097* 0.3402 0.9719**
[0.2719] [0.1160] [0.3157] [0.6489] [0.3061] [0.4326]
Percent of neighboring states adopted (lagged one year) 0.0073 0.1723 -0.009 0.9747
[0.1404] [0.7885] [0.4505] [2.8315]
Percent of neighboring states adopted, squared (lagged one year) -0.3338 -0.6081
Year Indicator Variables Y Y Y Y Y Y
Observations 48 48 48 49 43 43
Number of states 12 12 12 13 11 11
Hansen statistic (P-value) 0.503 0.998 0.999 0.53 0.92 1
Arellano-Bond, AR(1) (P-value)
Arellano-Bond, AR(2) (P-value)
Table 3-9. Continued
Early Adopters (1st Quartile) Late Adopters (3rd Quartile)
(7)t (8) (9) (10)1 (11)? (12)
Log of deflated R&D expenditures (lagged two years) 0.7061*** 0.8472*** 0.8594*** 0.8317*** 0.8053*** 0.9209***
[0.0714] [0.0343] [0.0346] [0.0641] [0.0691] [0.0647]
Prudent Investor Indicator (lagged one year) 0.2822** -0.6323** -0.4427 -0.2947** -0.0068 0.2988
[0.1114] [0.3033] [0.7028] [0.1168] [0.2409] [0.3620]
Percent of neighboring states adopted (lagged one year) 0.0924 -0.6519 -0.7924 -1.7842*
[0.3670] [2.4100] [1.7390] [1.0233]
Percent of neighboring states adopted, squared (lagged one year) 0.4223 1.7835
Year Indicator Variables Y Y Y Y Y Y
Observations 48 48 48 45 39 39
Number of states 12 12 12 12 10 10
Hansen statistic (P-value) 0.842 0.998 1.000 0.477 0.998 1.000
Arellano-Bond, AR(1) (P-value)
Arellano-Bond, AR(2) (P-value)
*F p<0.10, ** p<0.05, *** p<0.01. Robust standard errors are in brackets. Regressions were run using the Blundell-Bond dynamic
panel estimator. The symbol 'tf' indicates a two-step robust GMM estimation. All other regressions are robust one-step regressions.
Alaska and Hawaii are dropped when including data on neighboring states. The regressions for the Late Adopters (4th Quartile) are
not reported because there is no change in the prudent investor indicator variable. Late Adopters (3rd Quartile) were included instead.
Top R&D States (1st Quartile): Illinois, New York, Washington, California, Connecticut, Minnesota, New Jersey, Massachusetts,
Ohio, Pennsylvania, Michigan, and Texas. Bottom R&D States (4th Quartile): South Dakota, Montana, West Virginia, Nebraska,
Arkansas, Hawaii, Maine, North Dakota, Alaska, Vermont, Washington,
Table 3-10. Estimates of the Impact on Citation Weighted Patent Counts in a State
Top Patenting States (1st Quartile) Bottom Patenting States (4th Quartile)
(1)t (2) (3)t (4)t (5)t (6)t
Citation Weighted Patent Counts (lagged one year) 0.8221*** 0.8847*** -0.2708 0.6544* 0.0859 0.0523
[0.0518] [0.0059] [0.9829] [0.3778] [0.3800] [0.4678]
Prudent Investor Indicator (lagged one year) -1960.1743 -1981.2384 -146977.1501 127.4741 260.3926* 298.8881
[1,985.7316] [1,572.3731] [124,809.3084] [746.4490] [155.6210] [206.0809]
Percent of neighboring states adopted (lagged one year) -3101.5074 112319.2625 232.87 -254.71
[2,345.8993] [92,315.6510] [226.8410] [1,303.2966]
Percent of neighboring states adopted, squared (lagged one year) -391398.7295 512.31
Year Indicator Variables Y Y Y Y Y Y
Observations 120 120 120 130 110 110
Number of states 12 12 12 13 11 11
Hansen statistic (P-value) 1.000 1.000 1.000 1.000 1.000 1.000
Arellano-Bond, AR(1) (P-value) 0.074 0.134 0.539 0.061 0.079 0.163
Arellano-Bond, AR(2) (P-value) 0.12 0.201 0.794 0.858 0.097 0.292
Table 3-10. Continued
Early Adopters (1st Quartile) Late Adopters (3th Quartile)
(7) (8) (9) (10) (11) (12)
Citation Weighted Patent Counts (lagged one year) 0.8162*** 0.8236*** 0.8260*** 0.8535*** 0.8636*** 0.8605***
[0.0146] [0.0159] [0.0189] [0.0047] [0.0137] [0.0084]
Prudent Investor Indicator (lagged one year) -588.0025 -625.1823 -645.9265 -486.5116 -925.3682 -963.4427
[661.2821] [628.0003] [613.8107] [759.1179] [860.0614] [894.0843]
Percent of neighboring states adopted (lagged one year) 1,194.2446*** 2366.0929 54.445 1042.0041
[400.3876] [1,448.2099] [790.8376] [1,514.9252]
Percent of neighboring states adopted, squared (lagged one year) -1458.6684 -1,076.40
Year Indicator Variables Y Y Y Y Y Y
Observations 120 120 120 120 100 100
Number of states 12 12 12 12 10 10
Hansen statistic (P-value) 1.000 1.000 1.000 1.000 1.000 1.000
Arellano-Bond, AR(1) (P-value) 0.176 0.181 0.173 0.118 0.123 0.127
Arellano-Bond, AR(2) (P-value) 0.233 0.246 0.241 0.225 0.213 0.213
*F p<0.10, ** p<0.05, *** p<0.01. Robust standard errors are in brackets. Regressions were run using the Blundell-Bond dynamic
panel estimator. The symbol 'tf' indicates a two-step robust GMM estimation. All other regressions are robust one-step regressions.
The regressions for the Late Adopters (4th Quartile) are not reported because there is no change in the prudent investor indicator
variable. Late Adopters (3rd Quartile) were included instead. Alaska and Hawaii are dropped when including data on neighboring
states. Top Patenting States (1st Quartile): Illinois, Florida, New York, California, Connecticut, Minnesota, New Jersey,
Massachusetts, Ohio, Pennsylvania, Michigan, and Texas. Bottom Patenting States (4th Quartile): South Dakota, Montana, West
Virginia, Nebraska, Arkansas, Hawaii, Maine, North Dakota, Alaska, Washington, D.C., Wyoming, and Nevada. Early Adopters (1st
Quartile): Delaware, Illinois, Florida, Maryland, New York, South Dakota, Colorado, New Mexico, Oregon, Utah, Washington, and
Oklahoma. Late Adopters (3rd Quartile): Arkansas, Minnesota, Hawaii, New Jersey, Alaska, Vermont, Washington, D.C.,
Massachusetts, New Hampshire, Ohio, Indiana, and Wyoming.
DO PATENT ATTORNEYS MATTER?
The recent patent infringement lawsuit between Research In Motion and NTP has
increased public awareness of problems at the United States Patent and Trademark Office
(USPTO) and the various issues that currently burden patent policy in this country.
While the general public has only recently become aware of the problems, economists
have studied these issues for several years, with research on the topic generally falling
under two categories. The first group of research analyzes the impact of legal changes
that occurred during the 1980's and 1990's (see Jaffe (2000) for a good overview). The
second group examines institutional problems within the USPTO, specifically the
incentives for and effectiveness of the corps of examiners. One common element ignored
in the literature is the role of the patent attorney in the patent approval process.
While economists ignore lawyers, inventors tend to rely upon them during the
application and examination stages (to say nothing of during litigation). In assembling an
application, lawyers may search for prior art, make citations, and write claims to assure
that applications comply with USPTO regulations and to increase the odds a patent will
be granted. They are also the primary contact during the application process and may
meet with examiners, in person, to discuss an application. If inventors are right, and
lawyers provide an invaluable service, then economists have ignored an important
component of the patenting process. If lawyers are as unimportant as implied by
previous research, then inventors are wasting thousands of dollars by hiring one. Thus, to
fully understand the economics of patents it is important to be able to answer the
question, "Do patent attorneys matter?"
This chapter is the first to address the question by investigating how lawyer
characteristics affect a patent application's grant lag--the number of days between the
date an application is filed and the date a patent is granted. To do so, a unique dataset
that contains the names of lawyers representing the patent application was collected.
Additional lawyer characteristics, such as experience and busyness, were calculated to
help quantify and isolate the impact. The empirical analysis shows that lawyers do affect
a patent' s grant lag, but the magnitude of the impact differs by an invention' s technology.
Another finding is that the estimated impact of examiner experience on the grant lag is
different when lawyer characteristics are included than when they are excluded. This is
an important discovery since previous studies have suggested that differences in
examiner experience affect the quality of patents granted.
The rest of the chapter is as follows. The next section motivates the study with a
review of the relevant literature. In the third section, the data and variables are described.
The fourth section presents the empirical methodology and the fifth section discusses the
findings of the research. The final section concludes.
4.2 Literature Review
4.2.1 Theoretical Models
Theoretical models such as Segerstrom (1998), Denicolo (1996) underscore that the
timing of an innovation or patent is important. Firms compete in R&D in order to
develop the next best quality of a product and are compensated with monopoly profits
until the next quality innovation is discovered. In patent races, such as these, it is
imperative that researchers be the first to discover a new quality, or else they cannot
recoup their investment in R&D. Denicolo's model is particularly intriguing in that it
estimates the optimal breadth and length of patent protection. While the length of patent
protection in the United States is statutory, the breadth can potentially be influenced by a
lawyer representing the patent application. To extend Denicolo's thinking by including
legal intermediaries is, in truth, stepping beyond the original model, since having
representation implies there is some sort of negotiation and asymmetric information.
These elements are not present in the current literature on patent races or even some
criticisms of the USPTO (e.g. Griliches (1989) and Merges (1999)). They are, however,
included in models of bargaining.
Bargaining games of alternating offers, such as those in Osborne and Rubinstein
(1990), also motivate the analysis and the importance of timing. In these models, two
players bargain over the allocation of a pie of finite size and negotiations continue until
an agreement is reached. Each player prefers to possess a larger piece of the pie, which
has adverse effects on the size of the pie possessed by his/her opponent. The "driving
force of the model" is that each player has strong preferences over how long it takes to
find a solution. In the full information case an agreement between the two players is
immediately reached. When one player is uninformed, the equilibrium may not occur
immediately since "each player may try to deduce from his opponent' s moves the private
information that the opponent possesses." (p. 91) Regardless of when the agreement is
reached, the uninformed player receives a smaller allocation of the pie compared to the
fully informed player.
With this understanding, it is possible to draw a complete analogy to patent
negotiation; where breadth of protection is the pie to be divided (breadth of zero implies a
rej ected patent) and the lawyer and examiner are the players. It is clear why the owner of
an invention (and therefore lawyer) may prefer a broader patent since they may be used
to deter competition (see Scotchmer (2004)), increase revenue from licensing (Kamien
(1992) and Scotchmer), or facilitate cross-licensing agreements (in the context of
semiconductors, see Hall and Ziedonis (2001)). What is not necessarily clear is why
examiners may care about the size of the pie, since they do not directly benefit from their
allocation. Examiners are agents for society and future inventors. They are therefore
obligated to withhold patent breadth from the current inventor for use by future
inventors.' If the lawyer is the fully informed party (or, by extension, the one with the
most skill) the pie will be divided in his or her favor and potentially done so quicker than
under the opposite scenario. This chapter tests the accuracy of both predictions.
4.2.2 Previous Empirical Analyses
Much of the existing empirical literature on U. S. patent policy deals with legal
issues and the various causes and consequences of increased patent litigation. Even
though this research considers the role of lawyers, literature on litigation is not helpful
since these studies consider events after a patent is granted rather than during the
examination process. Papers that consider institutional issues at the USPTO provide
significant guidance on the topic. Three of the most significant papers in this area are
Cockburn, Kortum, and Stern (2003, CKS), King (2003), and Popp, Juhl, and Johnson
SMerges (1999) argues that the incentives provided to examiners are not necessarily in line with the
demands of society.
2 MergeS (1999) criticizes the experience of examiners and suggests reforms but does not provide
supporting empirical evidence.
CKS (2003) explore how differences between patent examiners affect the quality of
granted patents. Their data is based on 182 patents that were examined by the Court of
Appeals for the Federal Circuit. In addition to patent data they collected information on
the examiners that processed the patent application. The examiner data includes tenure
with the USPTO, actual time spent evaluating a patent, size of examiner' s docket, and
total number of patents the examiner reviewed. They cautiously conclude that examiner
discretion, but not workload or examiner experience, may affect the court' s decision to
rule it invalid.
King (2003) uses USPTO Time and Activity Reports to explore the examination
process and patent quality. He finds that examiner hours and examiner actions have been
relatively constant over time, in spite of the increased number of patent applications
submitted to the USPTO. He also shows that other measures of examiner quality, such as
pendency, have declined. These findings reinforce the notion that patent examiners are
extremely important to the patenting process. Of particular interest for this essay is
King's finding that as the number of examiner actions per patent increase, so does the
probability of approval. As King puts it, "this is consistent with examiners interacting
with inventors to improve the application and resulting patent award, a manifestation of
examination quality." (p. 65) While this tacitly supports this essay's hypothesis, he is
cautious since poor inventions may also require additional help from examiners.
PJJ (2004) explore who is most affected by long grant lags, not the causes of the
grant lag. Controlling for characteristics of the application, assignee, country of origin,
and industry (among others) they find that the primary driver of grant lag is the
technological field of the invention. They find that patent applications from the
biotechnology and computer sectors are most likely to have a longer grant lag. Given
their interests on effect rather than causality, examiner characteristics are omitted from
These three empirical papers help to motivate the empirical methodology for this
research. Following PJJ's (2004) example, grant lag is included as the dependent
variable.3 In addition, some of the variables (e.g. country indicators, number of Eigures,
and number of drawing pages) used by PJJ are included in the analysis. Based on the
Endings of CKS (2003) and King (2003), this analysis includes measures of examiner
experience and busyness. These data and their sources are discussed in more detail in the
4.3 Sources and Descriptive Statistics
4.3.1 Data Sources
The primary source for patent data is the NBER patent citations data-file (Hall,
Jaffe, Trajtenberg, 2001) which has recently been updated by Bronwyn Hall (Hall). The
dataset was expanded to include information from the public-use Hiles made available by
the USPTO (U. S. Department of Commerce). The new information included the full
application date, number of drawing pages, number of Eigures, the name of the primary
examiner and assistant examiner (if present), and the names of lawyers) representing the
patent.4 The names of examiners and lawyers extracted from the USPTO files were not
consistently formatted. On some patents, the individual is identified by surname and first
3 Due to data limitations, the other dependent variables used in King (2003) and CKS (2003) could not be
4 IH Some cases the name of the law firm was listed on the patent, not the name of the lawyerss.
Unfortunately, due to mergers between firms and inconsistencies in how the law firm name was entered,
patents that list the law firm name are of limited usefulness.
and middle initials. Other patents may exclude the middle initial and print out the full
first name. Still others may include "Jr.," "III," "Esq.," or a maiden name. These
irregularities had to be corrected in order to properly estimate the contribution of lawyers
The USPTO requires that all attorneys and agents register with the Office of
Enrollment and Discipline (OED) before being allowed to prosecute a patent application.
Upon registration, lawyers are issued a registration number that is to be used on all
official correspondence with the USPTO. Since registration numbers are awarded only
once, it is possible to match the names that appear on a patent with those that appear on
the OED's register, regardless of format or typographical errors. A list of lawyer names
and registration numbers were compiled through various searches on the USPTO's
website.5 Names on a patent that could not be confidently matched with a registration
number were marked and not included in the analysis.
Compared to lawyers, examiner identification was slightly more difficult since the
USPTO does not have a list of registration numbers or other unique identifier publicly
available. As a result, identification of examiners can solely be based on the names
printed on the patent. To identify examiners, each unique primary examiner name was
assigned an identification number. A slight misspelling, e.g. inclusion of suffix or middle
5 A list of registered lawyers currently active is available for download from the OED's webpage. The list
of active registrants was expanded by searching the USPTO's Patent Application Information Retrieval
website (PAIR) and collecting registration numbers and names that do not appear in the active roster.
Registration numbers and names were also found via a document identifying attorneys and agents that did
not respond to an OED survey. The lowest known registration number-name pairing in the 13,388 and the
highest known is 56,991. The compiled roster includes 31,673 names and registration numbers. This set of
names and registration numbers were then matched to those names listed in the expanded patent dataset.
Care was taken to assure that matching was as accurate as possible. If the accuracy of a match was in
question due to misspellings, an additional search of PAIR was made. If a match could not be verified the
lawyer was not assigned a registration number. Lawyers with known duplicate names were not included in
initial, may cause several different identification numbers for the same lawyer. To
account for these issues, several different sets of examiner identification numbers were
created and each set differed by its sensitivity to typographical errors, use or ignorance of
"Jr." or "Sr.," or inclusion of middle initial. Regressions were run using both sets of IDs
and the results did not significantly change. The results presented here use the least
tolerant IDs. Assistant examiners were not similarly identified since not all patents have
an assistant examiner and, if they do, work done by assistant examiners is subj ect to
approval by the primary examiner. An indicator variable was used to denote the presence
of an assistant examiner working on the patent application.
4.3.2 Description of Variables
This research uses data on patents granted in either 1999 or 2000 to quantify the
impact that legal intermediaries have on the grant lag--the number of days between when
an application was filed and a patent was approved.6 Many other factors besides lawyer
characteristics may affect a patent' s grant lag. These include application, invention,
country, examiner, and other characteristics as described in Table 4-1. Of these,
application, invention, and country characteristics are those most commonly included in
previous research. The effect of the number of figures on the grant lag is ambiguous. In
some cases, more figures may imply more clarity, making the examination of a patent
quicker and easier. At the same time, more figures may increase the amount of work an
examiner has to do since he or she must assure that what is claimed in the patent matches
the figures and that the figures are clear. Similarly, the number of drawing pages may
6 In 1995, the United States changed the length of a patent from seventeen years after the grant of a patent
to a length of twenty years after the application date. Successful applications filed after June 8, 1995 were
awarded the twenty year term. Only patents that were applied after that date were included in the dataset to
avoid issues caused by this change.
increase or decrease the grant lag. More pages may mean more clarity, but they might
also require more work by the examiner. The number of claims should have a positive
impact on the grant lag. Each claim must be evaluated by the examiner in the context of
the invention, but also based on prior art. As the number of claims increases, so should
the amount of time the examiner has to spend on the patent. An added benefit to
including claims is that it may capture the complexity of an invention. A more complex
invention will likely have more claims, and thus lead to an increase in the grant lag. The
number of citations will likely also have a positive impact on the grant lag, since each
must be checked by the examiner.
The contribution of an invention to society or an industry may also play a role in
the time a patent is reviewed, however importance is difficult to quantify. Two measures,
a generality index and citations received, are included in the regressions to attempt to
capture variation in grant lag associated with importance. Each was included in Hall's
updated patent citations dataset (Hall). The generality index is calculated as a
Herfindahl-type index and evaluates how important the patent was to inventions of other
technologies. A high index implies that the patent was cited by patents from a broad
range of technologies. It is expected that patents with a higher generality index will have
a reduced grant lag since important inventions likely have little prior art to make
comparisons. An important invention may also be cited more by other patents. To also
account for an invention's importance the number of citations received from other patents
is also included in regressions.
Who owns a patent may also affect the grant lag since some firms may prefer
shorter lags than others; however, computational constraints prohibited the inclusion of
assignee indicator variables to fully capture the assignee effects. Another, less
burdensome measure included in regressions is the type of the assignee, i.e. company,
government, individual, or unassigned. Since companies are subject to competitive
pressure they may strongly prefer a shorter lag whereas a government may solely be
interested in receiving a patent and not necessarily the timing of a patent. Assignees may
also differ in their ability to pay a lawyer and it is likely that larger companies have more
money available to hire patent attorneys. Financial data for assignees was not available
but it is possible to account for assignee size by looking at the number of patents the
assignee has owned over time, and how large that number is compared to other firms.
Regressions include two indicator variables, Assignee Percentile (75) and Assignee
Percentile (90), to indicate if the assignee was in the 75th or the 90th percentile of all
assignees in terms of patent ownership, respectively. It is expected that the signs of these
variables will be negative. Since not all patents are assigned, an indicator variable is
included that equals one if the patent was assigned and zero otherwise. Another variable
included is an indicator of the grant year. This is included to capture general USPTO
effects as well as trends in patenting and innovation. Many international inventors Eile
through the World Intellectual Property Organization (WIPO) and applications are later
transferred to the USPTO. The date of application extracted from the patent files may
reflect the date it was Hiled with WIPO and not with the USPTO. Therefore, international
patents may have a longer grant lag than domestic patents. To account for these issues, a
set of indicator variables denoting the country of origin was included in regressions. In
addition to capturing measurement error they may also capture communication problems
between an examiner and the inventor.
Like CKS (2003) and King (2003), this essay includes examiner characteristics, but
there is an important difference between their measure of experience and the one used
here. In CKS and King, examiner experience is the actual number of patents each
examiner reviewed during his or her tenure at the USPTO. Here, examiner experience is
based on the number of patents the primary examiner reviewed since mid-1 996. Since
data prior to 1996 is not available, there is imprecision in the calculations of examiner
experience. For this reason only patents granted in 1999 or 2000 were included in
regressions.' It is assumed that, over time, experience estimates will reach a level
proportional to their true levels. Even if this is not true, experience prior to mid-1996 is
essentially fixed and should be captured in the set of examiner indicator variables
included in all regressions. The impact of experience on the grant lag is likely nonlinear.
Early in an examiner' s career there is significant learning by doing and they become
more efficient at processing patenting applications over time. However it is likely that,
beyond a certain point, experience of the examiner does not improve performance. To
account for this nonlinearity, an experience squared term was included in regressions.
It is possible that a large examiner workload will cause delays in the patent
approval process. In addition to examiner experience and examiner indicator variables,
regressions also include a measure of examiner busyness. Examiner busyness is a count
of the number of patent applications that are currently part of an examiner' s docket. The
count is for all patents granted from mid-1996 to 2002 that have overlapping application
dates and/or overlapping grant dates.
SPatents granted in 2001 and 2002 are not included because generality measures are biased. See Hall,
Jaffe, and Trajtenberg (2001, p. 21-23) for more details.
To analyze the impact of lawyers on patent applications this essay considers two
separate measures of lawyer influence. The first measure is an indicator variable that is
equal to one if there was a lawyer listed on the patent application. Here the presence, but
not the skill, of a lawyer is measured. This general measure of lawyer impact also allows
patents that list a law firm (as opposed to a just a lawyer) to be included in the
regressions. The cost associated with inclusiveness is that an indicator variable cannot
capture the effect of an individual lawyer' s skill. This shortcoming leads to the second
measure of lawyer influence-individual lawyer attributes.
Skill at patent prosecution can be both dynamic and static. Dynamic skill implies
acquired knowledge or learning by doing. Static skill can be the result of intelligence,
education, personality, or other characteristics that are time invariant and inherent to the
lawyer, or at least did not change after passing the OED examination. The second set of
regressions account for both static and dynamic lawyer attributes by including a set of
lawyer indicator variables, lawyer experience, and lawyer busyness. As with examiner
experience, lawyer experience may be nonlinear. Lawyer busyness and experience were
calculated in the same manner as examiner busyness and experience. Beyond a certain
point, additional experience may not make a lawyer more efficient at getting a patent
approved. Therefore regressions also include lawyer experience squared. Since these
characteristics are individual specific effects, patents that list the law firm were not
included in the analysis.
Patent applications often have more than one lawyer working on them. As such, it
would be difficult to completely disaggregate the impact of each individual lawyer on the
patent application. It is assumed that one of the lawyers is the lead lawyer who directs or
guides the work of the other lawyers listed on the patent. For the purposes of this study
the lead lawyer is either the first lawyer listed on the patent or the most experienced
lawyer out of the entire set of lawyers.8 Each is considered separately.
One factor that may be an important determinant of grant lag, but not yet been
discussed is the technology of an invention. As Cohen et al. (2000) and Cohen et al.
(2002) point out, different industries prefer different forms of intellectual property
protection.9 Technology type may also affect the impact of legal representation and
lawyer experience. To account for both grant lag and lawyer issues, separate regressions
are run for fourteen different technology subcategories. The subcategories are based on
the Patent Classification System as of December 31i, 1999 as presented in Appendix 1 of
Hall, Jaffe, and Trajtenberg (2001). Table 4-2 lists the fourteen selected subcategories
and the top three assignees for each subcategory for patents granted during the 1999-2000
window. The chosen technology subcategories intentionally cover a wide range of
products including golf equipment, semiconductors, and agricultural products. This
facilitates comparisons to the entire set of technologies.
Tables 4-3 and 4-4 provide summary statistics of selected independent variables.
The statistics are separated by technology subcategory to underscore the potential
differences across technologies. The information is based on patents where the identity
of either the first lawyer or most experienced lawyer is known. Note that there is
considerable variance in the number of lawyers, examiners, and patents for each
SThe assertion that the first lawyer is the lead lawyer is dubious if lawyer names are listed alphabetically
on patents. Visual inspection of the data does not suggest this pattern is consistently used across patents. It
makes sense that the most experienced lawyer be the lead lawyer since, as with most other positions, more
success typically leads to promotions.
9 Another example is Levin et al. (1987), however the survey occurred before significant reforms of U.S.
subcategory. Significant differences also exist between technology types for the average
number of Eigures and lawyer experience. In spite of these differences, there is
remarkable similarity in the average number of claims, average grant lag, and citations
received across technologies.
The number of international patent applications has increased recently and it is
necessary to control for impacts unique to the country of origin. Table 4-5 shows the
number of patents originating in each country for each technology subcategory. Not
surprisingly, Japan and Germany inventors appear most often in Semiconductor Devices
(Subcategory 46), Motors, Engines & Parts (53), and Optics (54). Table 4-6 presents the
correlations between independent variables and the dependent variable, grant lag, by
4.4 Empirical Methodology
4.4.1 Regression Specifications
Several different measures of lawyer impact are considered, each requiring a
different specification. The first measure of lawyer impact is simply an indicator of
having a patent lawyer listed on the patent. These regressions use the specification:
grantlag, = a,, + a,,Represent + a,,Represent x Genderality Index + X' a,, (4-1)
where Xis a vector of independent variables (see Table 4-1 for descriptions) and a~
is the vector of coefficients for those regressors, for patents in technology subcategory i.
This simple specification establishes a baseline understanding of lawyer impact without
being concerned with lawyer heterogeneity caused by differences in experience,
education, or skill. As a consequence of this, all patents for each technology type are
included in regressions, even if those patents list a law firm name as opposed to
individual lawyer names.
The baseline specification assumes all lawyers are on equal footing. In reality
lawyers may differ in a variety of ways including education, experience, personality,
nationality, and gender. All of these differences may, in some way, affect the ability of
the lawyer in representing a patent. To estimate the impact of these factors another set of
regressions are run for each of the fourteen industries. These regressions have the
grantlag = A, + AExperience + P,,Experience2 + P,,Lawyer Busyness
+ A4Experience Generality Index + Y' /
where Experience is the number of patents a lawyer previously represented, Lawyer
Busyness is the number of other patents the lawyer is representing at the same time, and
Y is the same vector of regressors as in X of Equation 4-1 plus a set of lawyer indicator
variables. B is the vector of coefficient estimates for Y. Since lawyer specific variables
are used, those patents that did not hire a lawyer and those that listed a law firm were not
included. To account for the two different definitions of lead lawyer, two separate sets of
regressions were run. The first set defined the lead lawyer as the first lawyer listed on the
patent. The second considered the most experienced lawyer as the lead lawyer.
4.4.2 Endogeneity of Lawyer Choice
Before moving forward, it is necessary to acknowledge the potential endogeneity of
lawyer characteristics. Lawyers are chosen by the applicant and are expected to achieve
an outcome consistent with the applicant' s desires. Thus the application' s treatment by
lawyer may depend on the applicant's expectations of the grant lag. If the applicant
expects a patent to take too long, he or she may choose a lawyer to reduce the grant lag.
Similarly, if the grant lag is too short (due to limited breadth or potential rej section of the
application) then an experienced lawyer will be hired to extend the grant lag. Under
either case, there may be an unknown variable guiding the choice of the lawyer and that
omitted variable may be biasing the coefficient estimates.10
One possible unobserved variable that could affect lawyer selection is the value of
the invention to the applicant. An invention may be of high value if it is of significant
importance to a product the inventor or assignee wishes to sell. It may also be of high
value if the patent would be an important piece of a patent portfolio or patent wall. If the
applicant values the patent highly, then the expected grant lag (regardless of how many
days it actually is) is too long, and they will hire an experienced lawyer to reduce the
grant lag. Under this scenario, the unobserved variable causes coefficient estimates to be
biased upwards. Another unobserved variable that may affect lawyer selection is the
invention's merit to be patented. The USPTO requires that inventions must be unique,
novel, and non-obvious for them to be patentable, all of which are unobservable to the
researcher. However, recent criticisms of the USPTO have revolved around the issuance
of patents of questionable validity. Most notably the NTP patents on wireless email.ll If
an invention is of questionable merit, then it likely requires more time and a better lawyer
to convince the examiner that it is a worthy invention. From this perspective, a shorter
"' An alternative argument that is less problematic for the analysis is that lawyer choice is endogenous but
lawyer characteristics, such as experience, are predetermined. The subtle distinction implies that
endogeneity bias does not exist and that the coefficient estimates presented here are correct.
11Other popular examples include the peanut butter and jelly sandwich (Patent No. 6,004,596: represented
by Vickers, Daniels, and Young), Method of Exercising a Cat (5,443,036; represented by Epstein, Edell &
Retzer), and Method of Swinging on a Swing (6,368,227; represented by Peter L. Olson, the 5 year old
expected grant lag would be associated with the choice of an experienced lawyer and
coefficient estimates would be biased downwards.
The common approach to dealing with endogeneity is to find instrumental variables
for all endogenous regressors (i.e. lawyer experience, experience squared, and the
complete set of lawyer indicator variables) and exploit the variation of the instrument to
get at the true contribution of the endogenous variables.
While the data on patent characteristics is rich, the data on lawyers is limited.
Aside from name, experience, and registration number, very little is known about a
particular lawyer. This paucity of data limits the set of potential instrumental variables.
In regressions not included in this chapter, assignee experience, dates for the OED
examination and the number of patents granted by the USPTO since a lawyer first
appeared on a patent were tested as potential instruments.12 Without exception, the
available instruments were invalid or weak. Since weak instruments may be just as bad
(or worse) than the problems of endogeneity, only OLS estimates are presented.
Even though OLS estimates may be biased, knowing the direction of the bias may
facilitate interpretation. In the preceding discussion two potential omitted variables,
value and merit, showed the bias may be positive or negative, respectively. However it is
likely that invention value is more powerful than an invention's merit. Hiring a lawyer
imposes an additional cost (about $10,000 according to Barton (2000)) on the inventor or
assignee. If these costs could not be offset by additional gains, no lawyer would be hired.
Arguably, high value inventions provide a better means to offset the cost of a lawyer than
meritless inventions. Therefore, it is assumed that the endogeneity bias, if it exists, is
'2 See Appendix C for a more thorough discussion of the instrumental variables considered.
positive and that the coefficients are conservative estimates of the true contribution of a
4.5.1 Estimated Impact of Lawyers
Table 4-7 presents the results of regressions using the specification in Equation 4-1.
Regressions that exclude the representation indicator are also presented in the table to
highlight the contribution of representation. All regressions include application,
invention, country, examiner, and other variables as described in Table 4-1. Note that the
first and second regressions in the table are for patents in all fourteen technology
subcategories. When considering all subcategories, the impact of representation is fairly
small; having a lawyer will decrease grant lag by approximately ten days. While the
impact is small, the estimate is highly statistically significant. Based on the technology
specific regressions, the driving force behind the result for all technologies appears to be
Nuclear and X-ray inventions. The presence of a legal intermediary for patents of this
technology apparently reduces the grant lag by about one month. Lawyers also appear to
have a statistically signification impact in Optics and Apparel & Textiles. Interestingly,
including the representation indicator does not significantly affect the coefficients on any
of the examiner relevant variables (examiner experience, examiner experiences squared,
and examiner busyness). Regressions with the representation indicator variable also
included an interaction term between representation and the generality index to test the
hypothesis that representation may increase the scope of patent protection. If the
hypothesis is true, the coefficient for this interaction term should be statistically
significant and positive. None of the selected industries confirm this relationship. The
signs and magnitudes of examiner relevant coefficients are consistent with findings from
CKS (2003) and King (2003).
Table 4-8 presents the regression results for patents where the identity of the first
lawyer is known but the regressions do not include lawyer effects. Table 4-9 presents the
results of regressions on the same set of patents but includes the experience, experience
squared, and busyness of the first lawyer listed on the application. Again, inclusion of
lawyer characteristics causes the exclusion of patents with no representation or those that
only list a law firm. When accounting for lawyer specific characteristics, the impact of
the lawyer is not the same across technology subcategories. The grant lag for patents in
biotechnology (subcategory 33), Miscellaneous Drug & Medical (39), Nuclear and X-
rays (44), Semiconductors (46), and Optics (54) appear to be affected by lawyer
experience. In these technologies, increases in lawyer experience will reduce the time it
takes for a patent to be approved. Since the estimates for lawyer experience squared are
generally positive and significant, there appears to be decreasing returns to lawyer
Regressions included an interaction term for lawyer experience and generality to
again test the hypothesis that experienced lawyers are fighting for more breadth. As
before, the coefficient for this term is predicted to be statistically significant and positive.
In Nuclear and X-ray, the only technology subcategory that the coefficient was
statistically significant, the sign is negative. This appears to rej ect one of the predictions
of bargaining theory as outlined in sub-section 4.2. 1, however caution is merited in its
interpretation. The generality index may indicate the importance of an invention to an
industry and therefore may be an imperfect measure of breadth. The importance of an
invention is unaffected by the lawyer as it is determined during the research and
development stage and not during the prosecution of a patent application. A better
measure of breadth is needed to fully (and accurately) test the predictions of bargaining
theory as they pertain to the patent approval process. Another Einding is that, as lawyers
become more busy, the grant lag increases. This is true in most industries, even those
that are not impacted by differences in lawyer experience.
The same specification was used on patents where the most experienced lawyer
was identified (see Tables 4-10 and 4-11). The results are very similar to those based on
first lawyer characteristics. A likely cause of the similarity is the large number of patents
with only one lawyer. In these cases, the characteristics of the first and most experienced
lawyers are the same. Be that as it may, the regressions using most experienced lawyers
further emphasizes that patent attorneys are significant in the patent approval process.
4.5.2 Examiner Experience and Generality
Regressions in the previous sub-section included an interaction between lawyer
experience and a patent' s generality index to test if, as bargaining theory would suggest,
more experienced patent attorneys attempt to expand the scope of patent protection. The
results indicate weak evidence against the prediction. Another, prediction of bargaining
theory is that experienced examiners would decrease the breadth of a patent. Thus, the
coefficient of an interaction term between examiner experience and the generality index
is predicted to have a negative sign. Recognizing the caveat that the generality index is a
flawed measure of patent breadth there still may be some information gained from testing
Regressions in this sub-section use the specification
grantlalg, = Tio + 7tl LowLawExp + 7i2 LowExamExp + 7ta Examiner Busyness
+ Ti4 LowLawExp* LowExamExp + Tis LowExamExp Generality Index + Y' }t (4-3)
The main difference between equations (4-3) and (4-2) is the treatment of experience.
An indicator of low experience is used in these regressions rather than a count of the
number of previous patents represented or reviewed. Lawyers (examiners) that appear on
less than 10 (104) patents--the median experience for all lawyers (examiners)--are
considered to be low experience. The median was based on the highest amount of
experience gained for each lawyer (examiner), through 2002, regardless of technology
type. In addition, to the examiner-generality interaction, lawyer and examiner experience
indicators are likewise interacted.
Tables 4-12 and 4-13 present the results of these regressions using the experience
of the first lawyer or most experienced lawyer, respectively. As before, low lawyer
experience increases the grant lag in most cases. The impact of examiner experience is
less clear. In some cases low experience increases the grant lag, while in other
technology subcategories, the grant lag is decreased. This is in contrast to the evidence in
the previous subsection where examiner experience clearly decreased the grant lag. The
coefficient of the interaction term between examiner experience and the generality index
is rarely statistically significant. In the one case that it is statistically significant
(biotechnology), the sign is the opposite of the prediction. This again provides some
indication that bargaining may not be taking place, however further investigation with a
better measure of patent breadth is warranted.
4.5.3 Examiner Effects
As noted earlier, several papers have been extremely critical of examiners and their
incentives. The criticisms have been motivated, in part, by showing that differences in
examiner experience are a factor in the patent approval process. The results of this
inquiry confirm these findings. However, without exception, the coefficients for
examiner experience are lower (more negative) when not controlling for lawyer
characteristics than when they are included in regressions. For nuclear, x-ray, and
semiconductor patents, the coefficient for examiner experience is nearly half as large
when lawyers are included. This further suggests that examiner differences are important
but that these differences are overstated when not accounting for lawyer effects. In other
words, previous research that does not include lawyer effects has an omitted variable
4.5.4 Experience as a Proxy for Quality
It might be reasonable to interpret lawyer experience as a proxy for quality since
low quality lawyers are unlikely to continue to attract business. On the other hand, high
quality lawyers will likely attract a large amount of business and have higher levels of
busyness. Tables 4-14 and 4-15 summarize the impact of the median lawyer based on
coefficient estimates from subsection 4.5.2. The impact of lawyer quality, as proxied by
experience, is significantly different across technology types. Semiconductor
technologies experience the largest reduction in the grant lag. There, lawyer quality is
associated with reductions in the grant lag by approximately five months. In Pipes &
Joints and Gas technologies, lawyer quality (experience) appears to increase the grant lag.
Based on the descriptive information provided in Tables 4-3 through 4-6, there is no
obvious explanation for the result. It is therefore likely that these technologies are
intrinsically different than the others considered in this analysis and may be interesting
case studies for future research.
The evidence in Tables 4-14 and 4-15 confirms the findings of Cohen et al. (2000)
and Cohen et al. (2002). In those papers, survey evidence shows that product innovations
are more likely to use patents as a means of protection than are process innovations.
They also find that some industries (e.g. machine tools, medical equipment, and drugs)
place more of an emphasis on patents than others (e.g. food, textiles, and
printing/publi shing). Cohen et al. (2000) also examine the possibility that several
different mechanisms are combined to appropriate intellectual property. Respondents
that indicated a high regard for patents also stressed the importance of lead time; further
evidence that timing and patents are important and that lawyers should be an important
input in the patent approval process.
This research is the first to consider the impact that patent attorneys and agents
have on patents. Using an original dataset, it was shown that lawyers do affect the
pendency of a patent application, although the impact depends on an invention's
technology. In addition to the results on lawyer impacts, the essay reconfirms and
extends previous work on patents. Specifically, results provide further evidence that
different industries value patents differently and examiner experience does affect the
patent process. Importantly, including lawyer characteristics in regressions decreases the
importance of examiner experience. This suggests that previous research that do not
account for lawyer effects may overstate the problems associated with inexperienced
While the results are promising, the endogeneity of lawyers tempers their
applicability. A structural model on the role of lawyers could account for this problem
and lead to further understanding of the relationship between lawyers and examiners.
Bargaining games, such as those presented in Osborne and Rubinstein (1990) would be a
good starting point for such a model. More detailed data on lawyer characteristics may
yield stronger instrumental variables that also may account for endogeneity. Better data
on experience could allow more years of data to be included in the analysis. Finally, this
research does not explore repeated interactions between lawyers and examiners nor does
it estimate any differences between attorneys and agents (i.e. those with or without legal
training). These would be interesting extensions to research on patent attorneys.
Table 4-1. Variable Descriptions
Other EPO Countries
Examiner Experience, squared
Lawyer Experience, squared
Number of Claims by Patent
Number of Pages for Figures included in file
Number of Figures included in file
Number of citations made to other patents
Number of citations received by other patents
Herfindahl-like index of generality
Patents from Japanese inventors
Patents from German inventors
Patents from French inventors
Patents from British inventors
Patents from Canadian inventors
Patents from other European Patent Office countries
Patents from other countries (not U.S.)
Number of patents reviewed by primary examiner
Square of Examiner experience
Number of other patents primary examiner is working
on at the same time
Indicator variable if assistant examiner is used
(1 if true, else 0)
Indicator equal to one if lawyer or law firm was listed
Number of patents represented by lawyer (either first
or most experienced) since mid-1996
Square of Lawyer Experience
Number of other patents the lawyer (first or most
experienced) is working on at the same time
Examiner binaries Set of indicator variables for Primary examiner
Lawyer binaries Set of indicator variables for lawyer
(either first or most experienced)
Year binaries Indicator for year patent was granted (2000=1)
Unassinged Whether or not the patent was assigned
Assignee Type Type of Assignee (individuals, corporations, etc.)
Assignee Percentile (75)a Indicator if assignee is in the 75th percentile of all
assignees, by patent ownership
Assignee Percentile (90)a Indicator if assignee is in the 90th percentile of all
assignees, by patent ownership
a Calculation based on all patents granted from 1976-2000
Table 4-2. Technology Sub categories Descriptions
Subcategory ID Subcateogory Description Largest Patent Holders by Category
11 Agriculture, Food, Textiles American Cyanamid Company, Procter & Gamble Company, and Ciba Specialty Chemicals
13 Gas Air Products And Chemicals, Inc., Boc Group, Inc., and Praxair Technology, Inc.
33 Biotechnology Incyte Pharmaceuticals, Inc., Novo Nordisk A/S, and Smithkline Beecham Corporation
39 Miscellaneous-Drug & Med St. Jude Medical, Inc., Sulzer Orthopedics Inc., and 3M Innovative Properties Company
42 Electrical Lighting Motorola, Inc., U.S. Philips Corporation, and Philips Electronics North America Corp.
44 Nuclear & X-rays General Electric Company, Siemens Aktiengesellschaft, and U.S. Philips Corporation
46 Semiconductor Devices Advanced Micro Devices, Inc., International Business Machines Corporation, & Taiwan
Semiconductor Manufacturing Co., Ltd.
53 Motors, Engines & Parts Caterpillar Inc., Robert Bosch Gmbh, and Ford Global Technologies, Inc.
54 Optics Eastman Kodak Company, Fuji Photo Optical Co. Ltd., and Xerox Corporation
62 Amusement Devices Callaway Golf Company, Mattel Inc., and Walker Digital, LLC
63 Apparel & Textile Lindauer Dornier Gesellschaft Mbh, Sipra Patententwicklungs- Und
Beteiligungsgesellschaft Mbh, and Zinser Textilmaschinen Gmbh
64 Earth Working & Wells Schlumberger Technology Corporation, Weatherford/Lamb, Inc., and Halliburton Energy
66 Heating Babcock & Wilcox Company, Eastman Kodak Company, and IBM Corporation
67 Pipes & Joints Caterpillar Inc., Dayco Products, Inc., and General Electric Company
Source: Hall, Jaffe, and Trajtenberg (2001)