Week 4: Memos - Who Invents and Innovates? #12
Comments
Artificial Intelligence, Innovation, and Growth in South KoreaSouth Korea offers a fascinating case for examining the intersection of artificial intelligence (AI), innovation, and economic growth. With a $2 billion government investment in AI R&D since 2019, South Korea has emphasized building large, interdisciplinary teams to address critical national challenges in healthcare, manufacturing, and education. These initiatives aim to position the nation as a global leader in AI innovation. However, this strategy also highlights underlying tensions between team dynamics, innovation potential, and long-term growth—concepts central to this week’s readings. Large, hierarchical AI projects, such as the AI-driven national cancer detection initiative, focus on refining existing solutions through well-coordinated, top-down leadership. While this approach ensures immediate technological advancements and short-term productivity, it often lacks the flexibility required for disruptive innovation. This aligns with findings from Wu et al. (2019) and Xu et al. (2022), which show that larger, hierarchical teams excel at incremental development but struggle to disrupt established paradigms. Conversely, smaller, flatter teams within South Korea’s startup ecosystem have demonstrated significant potential for disruptive innovation. For instance, startups developing language models tailored to the Korean context have successfully challenged global players like OpenAI by leveraging local expertise and novel ideas. These examples underscore the importance of fostering diversity in team structures to balance the complementary strengths of disruption and development. To conceptualize this dynamic, I propose the following stylized equation: 
Here, the equation represents innovation output (I), which includes both disruptive contributions (novel ideas) and development-focused improvements (incremental advancements). Team hierarchy (H) and team size (S) are the critical variables in this framework. Flat teams (when H approaches 0) excel at generating disruptive, novel outputs. In contrast, hierarchical teams (when H approaches 1) with larger sizes focus on refining existing knowledge. The parameters (alpha, beta, gamma, delta) determine the relative importance of these factors in different contexts. Empirical evidence from South Korea’s AI patents (2015–2023) supports this framework. Smaller, flatter teams consistently score higher on disruption, introducing more innovative ideas, while larger, hierarchical teams dominate development-focused metrics by optimizing existing systems. This framework provides a conceptual tool for South Korean policymakers to evaluate team structures and innovation outcomes in AI projects. By fostering smaller, flat teams for early-stage breakthroughs while maintaining larger teams for development, South Korea can achieve sustainable AI-driven growth and innovation. |
|
Memo: The Role of Financial Aid in Addressing Talent Misallocation The reading emphasizes how talent misallocation leads to untapped human potential, particularly when gifted but low-income individuals lack the resources to reach their full potential or pursue advanced degrees like PhDs. To dig deeper into this issue, I analyzed the relationship between financial aid, measured by the percentage of Pell Grant recipients (PCTPELL), and average SAT scores at U.S. colleges. I used the Department of Education’s College Scorecard data for this analysis. Although SAT scores aren’t a perfect measure of talent, they offer a decent starting point for exploring how financial aid ties into academic preparation. The results were surprising. Schools with higher percentages of Pell Grant recipients tend to have lower average SAT scores. For instance, colleges in the 0-20% Pell Grant range have average SAT scores above 600, while those in the 80-100% range average closer to 500. At first glance, this might suggest that talented students are mostly found at wealthier schools with lower Pell Grant percentages. But I believe that conclusion oversimplifies what’s really going on. The graph highlights how financial aid often comes too late to effectively address the issue of talent misallocation in the U.S. college system. As we know, the reality is that SAT scores are deeply influenced by socioeconomic privilege. Students from wealthier families often have access to better-funded schools, private SAT tutors, and stable academic environments—all of which can inflate scores. On the flip side, students from low-income families—those most likely to attend high-Pell Grant institutions—often face poorly funded schools, fewer advanced courses, and limited access to test prep. These disadvantages don’t reflect a lack of talent but rather the inequities these students have to overcome. So, the relationship between financial aid and SAT scores probably says more about systemic barriers than about where talent is actually concentrated. This analysis makes it clear that financial aid alone isn’t enough to fix the misallocation of talent. While Pell Grants help open doors, systemic inequalities in K-12 education and test preparation still hold students back. To truly address this issue, policymakers and colleges need to invest in early education, make school funding more equitable, and rethink admissions processes to value more than just test scores. By doing so, we can start to unlock the potential of students who’ve been overlooked for far too long. 
|
|
Monetary costs role in misallocation of Talent and its role on aggregate Economic Growth Given President Trumps recent executive order pausing all federal loans and grants indefinitely I wanted to look at the effect of loan availability and other factors on misallocation of talent. Occupational preferences accounted for a relatively small share of growth in U.S. GDP per person between 1960 and 2010. According to one of the references from the premier reading by Chang Hai Tseh, declining barriers to accumulating human capital were significantly more influential than declining labor market discrimination. Specifically, reduced obstacles to human capital accumulation explained 36% of GDP per person growth, compared to only 8% attributed to reduced labor market discrimination. In contrast, changes in group-specific occupational preferences contributed minimally to economic growth during this period.
The analysis highlights that reducing barriers to accumulating human capital has had a profound impact on economic growth, contributing significantly more to U.S. GDP per person than efforts to address labor market discrimination. The implication is clear: addressing misallocation of talent caused by inadequate access to education and training opportunities can unlock substantial economic potential. Below are key recommendations based on this analysis: Given the pivotal role of human capital in economic growth, policies should aim to improve access to student loans, particularly for underrepresented and socioeconomically disadvantaged groups. Ensuring low-interest rates and expanding loan forgiveness programs for critical industries can incentivize investments in education. Socio-economic background significantly affects educational and occupational outcomes. Programs that combine financial support with mentorship, career counseling, and access to networks can help address systemic inequalities and create a more level playing field |
What a ban on non-competes could mean for inventor distribution across small and large firmsThe FTC proposed a rule in 2023 to ban most non-compete clauses in employment agreements. The rule was finalized in April 2024, but a Texas court struck it down in August 2024. In this memo I want to A) identify the distribution of non-competes over inventor age and what this might mean for the potential gains from abolishing non-competes and B) identify how non-competes can act as a potential lever for policy-makers to distribute inventors between small and large firms. We can expect non-competes to be distributed in several different ways with regards to inventor age. One can either a uniform distribution, such that non-competes effect the young and old equally; a skew towards either the young or old who tend to be less productive and might have less bargaining power as a result of their inexperience or past (lack of) success; or a concentration on “middle aged” inventors; as a strategic measure to target the most productive class. Since inventor productivity is not uniform across age groups, each of these scenarios would imply a different impact on innovation and the potential gains from abolishing non-competes.
The above figure shows both non-compete incidence by age and patent incidence by age. The incidence of non-competes closely tracks productivity, pointing to the fact that inventors in their prime could be being “hoarded” by larger firms. This sort of behavior is also noted in (Akcigit, Goldschlag 2023) where it is observed firms look to hire inventors and shelve their innovations to avoid displacement. This distribution in particular suggests that there may be significant gains in innovation that can be made by reducing/banning non-competes as it would free up the most productive portion of inventors to smaller firms. I now want to consider how non-competes might affect the distribution of inventors, in particular how they skew distribution towards larger firms. Assume an initial distribution of inventors across small and large firms. Consider the scenario where inventors seek to move jobs. In each time period, Non-competes are generally more common among larger firms. The costs of enforcement and drafting of non-competes generally make them inaccessible for smaller firms. Larger firms are also able to offset the perceived inconvenience of non-competes through cash compensation and other worker amenities, allowing them to pursue more restrictive policies. This characteristic is reflected in the fact that the probability of moving from larger to smaller firms is The setup implies the update rule for the proportion of workers at large firms is: Simplifying, we obtain: Solving for equilibrium: A ban would collapse |
|
The demographics of inventors: |
Does YC prefer in-person startups more than remote ones?I don't know!!! (probably, but that's not from the data)!!I'm not sure what types of conclusions (and the strengths of those conclusions can be said about this data) but maybe it's at least interesting to look at. The paper Remote collaboration fuses fewer breakthrough ideas suggests that disruptive ideas are less likely to come from remote teams than they are from in-person teams. With the mission of Y Combinator being to invest in disruptive teams, I was wondering if there was some data we could see that suggested confirmed they had a similar belief by looking at the percentage of fully or partly remote companies in their portfolio, using COVID and their transition to an online program between W20 and W22 as some signal of preference versus circumstance. That is, during COVID, most teams were likely forced to be remote, and so we would assume the percentage of remote teams YC would accept to be higher as a result of a changing population. However, after COVID, when teams had the choice of being remote or not, we might assume that YC can be more selective of whether to invest in remote teams or not. Observing this trend of a population shift, we can see if YC's preferences for remote teams remains the same or changes. Notes: (1) light orange represents remote batches, dark orange represents non-remote batches. (2) I grouped F24 and W25 as F24 switched to a quarterly model instead of a bi-annual one; F24 had 19% Any / Batch, whereas W25 had 13%.
From the graphs, we see that YC invested in less remote (both partly or remote) after they resumed in person programming. However, concluding whether these changes came from a population shift (companies no longer organized themselves online) or due to YC's preferences for in-person teams doesn't seem intellectually honest. Perhaps an argument can be said that both groups recognize the benefit of working in person, and thus the changes we observe are a result of founders changing the population makeup and YC being selective. But again, let's not draw any conclusions except that the percentage of remote companies, both partial and full, has been decreasing. It seems like we would need access to remote / in person data for all applications to a batch, both accepted and rejected companies, to begin thinking about drawing conclusions on preferences. Obviously there's a lot to be said about confounding variables / noise in this data. There's also a lot to be said about the limitations of this data.
From the directory and in the history of YC, YC considers itself to have 91 top companies, of which 56 are partly remote, and 14 are fully remote. A surface level reading might suggest that it is then okay to be at least partly remote. However, it's again unclear when these remote policies were implemented -- as Remote collaboration fuses fewer breakthrough ideas suggests, implementing remote policies might be productive for the development that happens after destruction. While we can't really draw conclusions about YC's preferences from the data, I think from testimonial, YC largely likes teams that are in-person. |
PhD Dissertations and Inventor-Sector RepresentationIn Tapping into Talent, Akcigit et al. discuss the role of government support toward academic PhD programs to stimulate innovation, such as by imposing minimum quotas for the number of PhD students that a university must accept. It’s clear that earning a PhD makes someone significantly more likely to invent. Just by the increased number of students exposed to PhD-level education following government support, there is a higher probability for some to innovate. However, I am curious about the efficiency of PhD programs in translating a student’s PhD degree into industry work––and specifically within patent-intensive fields––as opposed to academia. If there was a strong affinity between the distribution of research topics for a graduating PhD class and the distribution of young inventors across related sectors, then this would suggest that PhD programs are “efficient.” If instead the composition of research topics is not translated closely to young inventor rates across sectors, then there could be “inefficiency” in PhD programs only in the context of “invention”––and potentially better places to invest government support if the only goal is to bolster patent production. One way I wanted to investigate this relationship was to use PhD dissertations as an indicator of the research areas representing a university’s graduating PhD class. Ideally, I could collect all PhD dissertations produced within a group of research-intensive universities over the past ten years, and then for each author of a dissertation, determine if they produced at least one patent in a similar sector. Unfortunately, searching a patent database for thousands of dissertation authors is expensive for any accessible API. Instead, I gathered all PhD dissertations from a group of 23 R1 (very high research activity) universities over the past 5 years (> 40,000) using ProQuest. For each dissertation, I extracted a series of words/phrases about its subject (e.g. dissertation on drug responses could map to “immunology”) and accumulated the frequencies for the most common aggregate subjects. 
According to Measuring the Characteristics and Employment Dynamics of US Inventors, young (under 35) inventors maintain the highest representation share in the Information sector, followed by similar sector representation across FIRE and Prof/Sci/Tech Services. High representation in Information and Prof/Sci/Tech services matches closely with the high volume of PhD dissertations in computer science and artificial intelligence; these subjects comprise ~15% of graduating PhDs and have strong affinities to technology-intensive careers. Likewise, over 9% of dissertations discuss chemistry, which is applicable across diverse sectors. What’s more interesting is the high representation (~21.3%) of dissertations across topics in biology that aren’t correlated with high young inventor representation in the Health Care sector. In fact, healthcare represents the lowest share of young inventors. These PhD grads may be pursuing different careers in biology research that are productive in their own right––but not patent-producing. Non-targeted government support into PhD programs likely would not change this distribution. Furthermore, the dissertation distributions look different for different R1 universities––some institutions tend to produce more dissertations toward academia (e.g. UChicago), whereas others are more invention-relevant (e.g. MIT, computer science). This ultimately suggests that if a government wants to bolster inventors among other forms of innovation, they could identify 1) which universities have historically been successful in producing inventors, or 2) whether the education sector can be complemented in other ways beyond blanket PhD support. |
Differentiating the Demographics of Male vs. Female InventorsMotivation / Guiding Questions Our readings and lectures this week focused on who becomes inventors. Professor Akcigit introduced a model in class that teases out relationships between an inventor and demographic factors like father's education, family wealth, education level, etc. One of our readings this week focused on the tendency for women inventors to invent solutions catered for women, and that many promising female-focused discoveries have yet to be commercialized because women are less likely to obtain patents. My question comes from the intersection of these inquiries - how do the demographic backgrounds of the most successful inventors differ when separated by gender? Do female inventors tend to have different socioeconomic and educational backgrounds compared to their male counterparts? Do they have more privileged backgrounds (perhaps to atone for gender barriers and inequal access to opportunities) or fewer credentials? Methodology Education Level (High School = 0, College = 1, PhD = 2) Results
Before conclusions are drawn, a few caveats of this study should be noted.
The analysis revealed that, on average, female inventors had higher cumulative scores than male inventors, suggesting that women who become inventors often come from more privileged backgrounds—whether in terms of education, wealth, or parental influence—compared to men in the same field. One possible explanation for this is that women in male-dominated industries often require greater credentials or support to achieve similar professional success, perhaps due to barriers to opportunities and gender stereotypes, like suggested in our last reading this week (i.e. difficulty in obtaining patents). The scatter plot highlights this pattern, with female inventors being more concentrated in higher-score ranges, whereas male inventors exhibited a wider spread across lower-score categories. This result can be considered hand in hand with a study by the Russell Sage Foundation Journal of the Social Sciences reporting that the number of female bachelor's degrees holders have surpassed men since the mid-1980s, but this seemingly equal educational attainment is not continued when measuring professional advancement. Considerations for a Larger Study
|
Assessing the Causal Impact of Free Community College on InnovationInspired by the Regression Discontinuity Design discussed in class on Tuesday, in this week’s memo I would like to propose a similar experimental design investing the causal impact of community college subsidy programs on individual innovation. Context To the surprise of many, nearly half of college students are enrolled in community colleges. According to the Community College Research Center, community colleges enrolled 8.9 million students in the 2020-21 academic year, representing 41% of undergraduates. Over the last decade, hundreds of college promise programs were launched as a potential pathway for increasing equitable access to education for students across the United States. Today there are over 250 programs in 50 states with 47 programs administered and sponsored by state governments. Given the foundational importance of community colleges in broadening access to universal education, does it also positively impact innovation dynamics and entrepreneurship? Specifically, I would like to focus on the case study of the Star Scholarship Project in Chicago, a promise program launched in 2014 providing free tuition for the City Colleges of Chicago (CCC). This scholarship was available to all Chicago public high school graduates who met the following performance-based eligibility requirements:
What is the causal relationship between eligibility for the Star Scholarship and the probability that a student will become in inventor in the future? To answer this question, I propose a Regression Discontinuity (RD) design to estimate the local average treatment effect of access to Star Scholarship funds on the likelihood of becoming an innovator in the future. For the purposes of this exercise, we are not differentiating between transformative and subsistence innovators. The primary measure of innovation is having one or more successful matching between student information and patents registered with the USPTO. Assumptions:
Implementation To implement this study, I attempt to use a local linear approach involving the below steps:
If the hypothesis holds true, we shall expect to see a hike in the likelihood of becoming an inventor for those students who were right around the threshold of a 3.0 GPA and thus have received the Star Scholarship. |
OpenAI vs. DeepSeekIn light of recent events, the comparison between OpenAI and DeepSeek provides a compelling case study in how team size influences innovation. OpenAI, with extensive funding and a sizable research team, exemplifies how large teams refine existing architectures and optimize performance through incremental advances. DeepSeek, with a smaller, more agile research team, has instead introduced a disruptive efficiency innovation in AI. This distinction aligns with the paper “Large teams develop and small teams disrupt science and technology,” which demonstrates that large teams develop and refine ideas while small teams disrupt by exploring unconventional approaches. Wu, Wang, and Evans argue that large research teams excel at optimizing established technologies rather than creating fundamentally new ones. OpenAI, now an established company with 4,500 employees, fits this model, having raised over $20bn to scale existing architectures and fine-tune performance by continually improving new versions of their models. The GPT model, for instance, has been improved dramatically from its GPT-1 version to its current GPT-4o version through structured refinements within its transformer architecture rather than by challenging core assumptions about efficiency. OpenAI’s reliance on large-scale compute infrastructure further reinforces this pattern, demonstrating how large teams prioritize stability, scalability, and commercial viability over radical innovation. In contrast, DeepSeek, with around 200 employees, follows the disruptive trajectory of smaller research teams. According to Wu, Wang, and Evans, small teams take greater risks, explore less conventional knowledge, and are more likely to introduce breakthroughs. The DeepSeek-R1 model exemplifies this by challenging the dominant cost structure of AI models rather than competing purely on scale. Instead of relying on extensive computational resources like OpenAI, DeepSeek leveraged alternative efficiency strategies, producing a model that maintains high performance at significantly lower costs than the comparable OpenAI o1 model. DeepSeek claims that the cost of training its R1 model was only $5.6mm, more than 80 times less than the speculated training cost of the o1 model. This represents not just an incremental improvement but a paradigm shift in AI accessibility and computational efficiency. DeepSeek-R1 disrupts not by achieving superior performance outright but by redefining efficiency expectations in AI research. While OpenAI’s o1 refines existing models for scalability, DeepSeek-R1 democratizes access to high-performance AI by significantly reducing computational demands. The input cost of DeepSeek-R1 is only $0.55 per million tokens, as opposed to $15 for OpenAI’s o1. This reflects the idea that large teams improve and scale while small teams disrupt by reimagining foundational constraints. Furthermore, the OpenAI–DeepSeek comparison underscores the importance of supporting smaller AI teams that drive disruptive breakthroughs. Policymakers should fund alternative AI research paths and ensure equitable compute access, preventing innovation from being dictated solely by large-scale players that optimize existing technologies. In summary, DeepSeek’s development of R1 illustrates how small teams can redefine efficiency standards in AI, while OpenAI’s o1 reflects the structured refinement characteristic of large teams. This case study reinforces the argument that large teams develop while small teams disrupt, highlighting the need to balance both approaches to sustain AI innovation. 
|
The Gender Gap in AI Research and Its Impact on InnovationArtificial Intelligence is rapidly transforming industries, but who is leading its development? I started thinking about this question after reading the article “Who do we invent for? Patents by women focus more on women’s health, but few women get to invent". Research suggests that women are significantly underrepresented in AI research, raising concerns about whether the field’s priorities adequately reflect diverse perspectives.
The graph above visualizes gender disparities in AI research, comparing the percentage of women as authors, co-authors, and single authors in AI vs. non-AI fields. The data, drawn from arXiv publications and summarized in a Nesta article titled "Gender Diversity in AI Research", highlights three key findings:
These trends suggest that gender disparities persist not only in participation but in who gets to define AI’s research agenda. The consequences of this gap extend beyond fairness and representation; they also shape the way AI systems are designed and deployed. To explore these disparities further, I think it would be interesting to investigate the impact of gender representation on AI research funding. The goal would be to see if male and female researchers receive comparable levels of funding and if certain institutional policies have influenced funding trends over time. One approach could be to analyze grant amounts awarded to male-led versus female-led research teams, controlling for factors like research field and institutional affiliation. It could also be valuable to examine whether women in AI have different funding success rates compared to their peers in other STEM fields. Another angle would be to explore how funding affects research outcomes. Are female-led projects underfunded relative to their scientific impact? Do they produce fewer publications or receive fewer citations due to resource constraints? Thus, also taking a look at whether financial disparities contribute to the broader gender gap in AI. This kind of analysis could help shed light on structural barriers in AI research funding and offer insights into policy solutions that promote greater equity in the field. Understanding these patterns could inform future efforts to create a more inclusive AI research ecosystem. |
Growth Dynamics in Education-Innovation Policy InteractionsDrawing from Akcigit, Pearce, and Prato's analysis of Danish data, I propose a Growth-Friction-Policy (GFP) equation that captures how talent allocation frictions mediate productivity growth: Where:
This equation integrates three key mechanisms that determine aggregate productivity growth. The talent transformation term The empirical evidence strongly supports this framework. In Denmark's baseline case, productivity growth averages 1.5% annually. A 10% R&D subsidy alone generates only half the predicted growth impact of traditional models (5% vs 10%). The GFP equation explains this through the interaction between financial frictions ( This framework delivers important policy implications for productivity growth. By explicitly modeling how financial frictions mediate education policy effectiveness, it suggests optimal policy mixes should shift toward education subsidies in societies with higher inequality or financial frictions. It also captures why growth impacts vary over time - R&D subsidies work through existing researchers while education policies take longer but potentially deliver greater returns through improved talent allocation. The GFP equation thus provides not only a quantitative model for understanding how education and innovation policies affect productivity growth but also suggests strategies for optimizing their combination based on a society's underlying frictions and inequality levels. It implies that the most effective growth strategies will carefully balance immediate innovation incentives with longer-term investments in talent allocation efficiency. |
R&D Expenditure within U.S UniversitiesAmerican universities are valuable engines of innovation that shape the future of technology and science in the United States. Together, these institutions spend roughly $75 billion each year on research—about 13% of the nation’s total R&D spending according to the George Bush Presidential Center. A lot of this funding supports work in STEM fields, making universities a cornerstone of the country’s innovation ecosystem. However upon looking up the R&D expenditure among schools, I found a worrying trend emerging:
The chart shows the trends in R&D expenditures among the top 10 U.S. universities in R&D expenditure from 2019 to 2023. Sourced from the National Science Foundation, the data highlights institutions that consistently lead in research spending. I think the most striking pattern is the noticeable drop between 2019 and 2020, likely due to the COVID-19 pandemic, with little recovery or further declines in subsequent years. Universities like Wisconsin-Madison saw some of the sharpest cuts, while places like Johns Hopkins and Stanford experienced steadier, ongoing reductions. These trends suggest challenges for universities in maintaining their research productivity and innovation capacity. While the pandemic clearly played a big role in this decline, I think other economic forces are also at work. During times of uncertainty, businesses tend to tighten their R&D budgets, which means less funding for collaborative research with universities. On top of that, federal funding for university research has been shrinking for decades—from covering 67% of R&D expenditures in 1979 to just 55% by 2022. As technology advances, the costs of research—whether for cutting-edge equipment, regulatory compliance, or competitive salaries to attract top talent—have risen sharply, leaving universities under mounting financial pressure. This drop in R&D spending is a problem for innovation. Universities do not just advance technology; they also produce the skilled people who make those advances possible. According to Measuring the Characteristics and Employment Dynamics of U.S. Inventors, PhDs are 20 times more likely than the average person to become inventors. Cutting R&D funding risks slowing the flow of talent into innovation-driven industries. It’s also troubling in light of broader trends—like inventors clustering in older, larger firms and becoming less likely to switch jobs or start new companies. This reduced dynamism in the innovation ecosystem might only get worse if early-career inventors lose the support they need from university research. Younger firms, known for producing high-impact patents, depend on talent fostered in strong academic research environments. To stay competitive on a global scale, U.S. universities need to stabilize or grow their R&D investments. This can be done through collaborations with industry or from finding alternative funding sources. Recent policy moves, like the proposed federal spending freeze earlier this year, highlight just how vulnerable federal research funding can be. Even though the freeze was temporarily blocked, the uncertainty it caused is a reminder that universities need to plan ahead to weather these kinds of financial challenges. |
Team Size and FieldsIn Remote collaboration fuses fewer breakthrough ideas, Lin et al. look at how innovation conducted in remote settings produces fewer disruptive innovations because without on-site communication, integrating pieces to produce disruptive technologies becomes more difficult. To do so, they generate a dataset of millions of papers and patents as well as data about collaboration distance, disruption score, and team size. Using this data, I wanted to further explore the properties of teams that innovate. In particular, I looked at how team sizes vary in different fields. To do so, I used the first 1,048,576 (I chose 1,048,576 for performance purposes; the full dataset (>4e6 patents) was too large) patents from their dataset on patents. I was then able to use the field data, which is encoded via the Cooperative Patent Classification (CPC) codes, and was able to calculate the average team size for each field. The results are summarized below:
ConclusionsThe graph, by and large, doesn't show significant trends in team size by field, but there are some insights we can gain from this graph.
|
|
This week’s readings highlight the tension between disruptive innovation, which challenges existing paradigms, and developmental innovation, which refines and extends existing knowledge. Team size plays a critical role in determining whether research is more disruptive or developmental. This memo presents a simple model to illustrate this tradeoff and offers insights into how scientific institutions and funding agencies can optimize research structures for innovation. Mathematical Model
We model disruptive innovation as: and developmental innovation as: where K = e^-0.03T represents decreasing novelty as team size increases. Implications for Policy and Research:
|
|
To the average layman, many of the findings in the paper Measuring the Characteristics and Employment Dynamics of U.S. Inventors would not come as a surprise. Inventors are largely older, whiter, and wealthier than the general American population. They exhibit a strong superstar effect, often producing only one or two patents, many of which are not particularly valuable. What might be more unexpected, however, are the negative trends in the employment dynamics of American inventors.
The above is a theoretical equation on factors that drive innovation in the world of inventors. This obviously super theoretical but here a description of the possible parameters . Entrepreneurship (E) boosts disruptive innovation but has been declining. Older inventors (A) tend to slow down innovation, making age a negative factor (α). Earnings inequality (W) is a mixed bag (β)—while top inventors drive big breakthroughs, inequality might discourage new talent. High-impact patents (C) thrive in young firms (F), but these firms are shrinking. Geographic concentration (G) helps innovation (γ) by clustering talent, though it can widen regional gaps. Job mobility (M) fuels knowledge sharing (δ), but declining mobility hurts this. Finally, outside forces (ϵ) might constitute some errors and variability. This code serves as the basis of a future study on what demographics are best for innovation spell check this |
|
Talent Dynamics: Constraints & Spillovers Talent allocation evolves over time, shaping innovation potential while interacting with institutional and cultural constraints. Understanding these dynamics is crucial for designing policies that maximize talent spillovers, reduce barriers to entry, and create sustainable, high-capacity innovation ecosystems. I seek to extend the innovation equation Talent-driven innovation is a function of resource availability R and diffusion effectiveness D, while recognizing the role of cultural resistance C and institutional constraints L in shaping talent allocation. This static formulation, however, I believe overlooks the systemic frictions that shape the long-term evolution of talent and innovation capacity. A fundamental limitation of the equation in its current form is its assumption that talent flows remain constant over time. In reality, talent accumulation is path dependent. Today’s allocation decisions influence future innovation capacity. This means that R and D are not exogenous but functions of past talent deployment: In economies where high-skill individuals continuously engage in innovative sectors, talent spillovers enhance future productivity, increasing the effective stock of resources R and improving diffusion mechanisms D. This suggests that the equation should be reformulated as: Here, While C and L are often considered exogenous constraints, they are in fact endogenously shaped by the outcomes of talent allocation. A system that effectively channels talent into high-productivity sectors can erode resistance over time, fostering cultural acceptance of risk-taking and entrepreneurship. However, if talent policies fail to provide inclusive pathways for participation, barriers to entry may rise, increasing cultural and institutional frictions. To model this, we extend the constraint terms as: This adjustment leads to a more dynamic equation form: If |
Can Adjusting Measures of Ability Reduce the Demographic Gap in Education?This week’s readings demonstrated the importance of innate ability or merit in innovation. Standardized testing is the classic way of measuring educational merit. Unfortunately, studies have found that standardized tests show ethnic bias. In NYC, entrance into the city’s specialized high schools is based on SHSAT scores. This system has been hailed as a “colorblind meritocracy” but Black and Latino students are tragically underrepresented in specialized high schools. In 2023, only 7 of 762 Stuyvesant admits were Black. This underrepresentation has largely been attributed to differences in educational cultural significance and access to preparation resources, which influence the choice of studying for the SHSAT. I found the following data: This data shows that while the tester distribution is representative of NYC demographic data, the offer rate is not. This suggests an inefficacy in developing and capturing talent. As studying for the SHSAT increases your score, I estimated the normal distribution parameters of scores per ethnicity group if there was no studying for the SHSAT. I use the No-Studying scenario as a measure of ability that is less correlated with access to preparation resources. I calculated the following: (A) Standard Deviation: (B) Probability of Studying: (C) Adjusted Mean – Mean if No Studying: I consulted an SHSAT prep vendor, and reasonably estimated that the average student studies for the SHSAT for 48 hours. (The max score on the SHSAT is 700.) I assumed the Overall Mean was the baseline score. Thus, I calculated the Adjusted Mean as: (D) Adjusted Cutoff: (E) Adjusted Offer Rate: I compared the No-Studying distributions with the Unadjusted distributions. My calculations: The results of this analysis suggest that the offer rate for Native American, Black, and Hispanic students would increase if no student was allowed to study for the SHSAT. As I measured the No-Studying scenario as a richer proxy for innate ability, this also suggests that the proportion of Native American, Black, and Hispanic students at specialized high schools would increase if the admissions process used more accurate measures of ability. |
|
“Measuring the Characteristics and Employment Dynamics of U.S. Inventors” by Akcigit and Goldschlag shared insights about many notable features of US inventors. One of them is that “Many inventors are foreign born. Over 30% of inventors are foreign born and we see a rise and fall of foreign born share among young inventors. China and India account for an increasing share of foreign born inventors, rising from 25% to 40% by 2016”. I propose using instrumental variables (IV) to further investigate the research findings on foreign born inventors: specifically analyzing the impact of immigration policies on innovation output. However, this simple OLS model likely suffers from endogeneity issues, as there may be unobserved factors affecting both the share of foreign-born inventors and innovation output. To address this, we can use an instrumental variable approach. The IV model would consist of two stages: Second Stage: For this IV approach to work, the instrument (I) must satisfy two key conditions:
To implement this model empirically, I would need:
By estimating this IV model, I could potentially identify the causal effect of foreign-born inventors on innovation output, providing stronger evidence for policy recommendations regarding immigration and innovation. |
|
The “Engineer Driving for Uber” Phenomena The term “engineer driving for Uber” became a familiar term across Brazil in the late 2010s as a comic way of describing the phenomenon of college graduates and highly specialized workers taking on blue-collar, low-wage jobs. This situation is associated with the placement of many incentives across higher education systems in the country to include low-income students, public school graduates, and ethnic minorities. These programs included quotas within top universities and housing assistance programs. Although these programs have had a positive impact on increasing the number of skilled workers in the country, the demand from private firms for skilled workers did not increase at a similar speed, generating a large gap between the number of qualified individuals and available positions. Analyzing data from the Brazilian National Household Sample Survey (PNAD) and the Central Register of Enterprises (CEMPRE), it was found that the growth in college-degree-holding individuals was over double that of the industry demand for individuals with that level of certification between the years of 2015 and 2019 (25.4% to 11.9%). 
In class and in papers, much of the discussion surrounding economic development is about the lack of specialized talent for the growth of a country, and policy suggestions are often based on educational incentives. However, this specific case study shows the importance of alignment between education spending and private market incentives. The lack thereof makes the investment in education less productive, as individuals with college degrees end up in jobs where their degrees are not being used, resulting in an oversupply of educated workers without corresponding opportunities in their fields. One major issue that arises from this mismatch is brain drain, as many of the highly skilled individuals who struggle to find jobs in their field seek opportunities abroad, leading to a loss of human capital that could’ve been used to drive economic growth.This can also lead to wage suppression, as more candidates compete for the same few skilled positions, driving down salaries, and disillusionment with higher education, as future generations may begin to question whether a college degree is worth the investment if it does not guarantee upward mobility. To address this issue, the country should reconsider its firm incentive strategy to allow for growth in demand to accompany the increasing supply of skilled workers. Lowering corporate taxes and reducing bureaucratic barriers for startups would encourage business expansion, ultimately creating more skilled jobs. Additionally, public-private partnerships could be leveraged to ensure universities align their curricula with market needs. Firms and industries should collaborate with educational institutions to offer practical training programs preparing students for job opportunities. In the long run, Brazil’s economic competitiveness depends on its ability to bridge the gap between education and market needs. Without strategic reforms, the cycle of misaligned education policies, underemployment, and economic inefficiency will persist. By creating an ecosystem that fosters innovation, incentivizes business growth, and aligns educational output with labor market demand, Brazil can turn its skilled labor surplus into an economic advantage rather than a systemic problem. |
|
This memo was thought after further reading about DeepSeek and Nvidia in my Trading class at UChicago. This article was one we looked at in particular. Formula |
A Talent Mobility Model for India’s Biotech RevolutionIndia’s biotechnology sector is one of the fastest-growing in the world, projected to reach $150 billion by 2025, driven by advancements in pharmaceuticals, bioinformatics, and healthcare innovation. With a strong foundation in vaccine production, generic drugs, and bio-agriculture, companies like Serum Institute of India, Biocon, and Bharat Biotech have global recognition. However, how mobile is India’s biotech talent? Understanding the factors that influence inventor mobility—whether they remain in large, established firms or transition to startups and research-driven ventures—can offer valuable insights for industry growth and policy planning. Based on the concepts discussed in the “Measuring the Characteristics and Employment Dynamics of U.S. Inventors" reading, I propose a logit model to examine mobility in the Indian biotech space, capturing how firm characteristics, inventor productivity, and time trends shape employment transitions.
Where:
The expected coefficients, with real-world examples, are as follows:
To test this hypothesis, we would collect employment and patent data from sources like LinkedIn India, the Indian Patent Office, and CMIE Prowess, linking inventors to job transitions, firm size, and patent citations. A logistic regression analysis would estimate how these factors drive mobility, providing insights for business leaders, investors, and policymakers. Currently, the model suggests that increasing funding for biotech startups through initiatives like Startup India and BIRAC can ensure that high-impact inventors do not remain locked in legacy firms. Additionally, reforms in clinical trial regulations and faster patent approvals could accelerate the commercialization of research, making it easier for inventors to transition between firms and roles. |
The Role of Immigration in Expanding the U.S. Inventor PipelineIntroductionInnovation-driven economic growth depends on a steady pipeline of inventors contributing to technological advancement. A key question in the literature is: Who gets to invent? Research shows that access to quality education, financial resources, and supportive networks significantly influences innovation potential. However, another critical yet underexplored dimension is the role of immigration in shaping the U.S. inventor landscape. Immigrant inventors contribute disproportionately to innovation, filing a significant share of high-impact patents. This memo investigates the extent to which immigration expands the inventor pipeline and assesses policy implications for fostering immigrant-driven innovation. Empirical Evidence on Immigration and InnovationData from the USPTO, NBER, and Census Bureau indicate that immigrant inventors play an outsized role in the U.S. innovation ecosystem. Key findings include:
To quantify the impact of immigration on innovation, I propose the following regression model: Preliminary estimates suggest that immigrant inventors file 40-50% more patents than native-born inventors in STEM fields, even after controlling for education and funding. Key Findings and Interpretation
Policy and Research Implications
ConclusionImmigrant inventors are a crucial driver of U.S. innovation, yet visa restrictions and policy inefficiencies hinder their full potential. Expanding immigration pathways, reducing bureaucratic frictions, and investing in STEM talent development can ensure the U.S. remains at the forefront of global technological advancement. |
|
How does Inventor Education impact business success? The “Measuring the Characteristics and Employment Dynamics of U.S. Inventors” paper shows that inventors with PhDs are more likely to become active innovators in the U.S. But this led me to wonder if this educational advantage translates into actual business success. So using data from over 6,000 venture-backed companies, I analyzed whether different types of educational credentials really make a difference in startup success. The results are pretty interesting. PhDs aren't just more likely to become inventors– they're actually significantly more successful when they do. CEOs with doctoral degrees achieve success rates of 34.5%, compared to just 23.3% for those without. This 11.2 percentage point difference is fairly large, especially considering that only about 12% of CEOs in the sample have PhDs. Another element of my analysis which is interesting is how the different types of advanced degrees contribute to success. While MBAs are the most common advanced degree in the dataset (held by 21.6% of CEOs), they only seem to provide a modest boost to success rates. And Master's degrees show almost no impact at all. This might suggest that the specialized knowledge gained in a PhD program might be more valuable for innovation than broader educational credentials. 
(Success defined as whether inventors startup had annualized return > 25%) Another interesting thing is that more degrees aren't necessarily better. While having one advanced degree (Beyond undergrad) improves success rates from 22.1% to 28.3%, adding more degrees doesn't help and might even hurt. CEOs with three degrees actually show lower success rates, though this group is quite small in the sample (only 10). So while Akcigit and Goldschlag show that PhDs are more likely to become inventors, the analysis suggests they're also more likely to succeed when they do. But the data shows that the path to innovation success isn't just about collecting degrees, but rather about getting the right kind of education for the challenges of leading an innovative company. |
Singapore's Graduate Workforce: Skill InequalitiesOver the past decade, Singapore has significantly expanded its graduate workforce. The share of university-educated workers grew from 26.7% in 2009 to 37.5% in 2019, while jobs requiring a degree also increased from 29% to 38% between 2013 and 2017. On the surface, this suggests that higher education growth has kept pace with labor demand. However, a closer look reveals a lingering issue—many graduates are not in jobs that truly require their qualifications, leading to underemployment and wage gaps. 
A key distinction lies between task-warranted and task-unwarranted jobs. Task-warranted jobs genuinely require a degree because they involve specialized skills and advanced knowledge. Task-unwarranted jobs, on the other hand, may require a degree for hiring purposes but do not necessarily demand university-level expertise in practice. The data shows that while more graduate jobs are available, many fall into the task-unwarranted category. This creates economic inefficiencies, as graduates in task-unwarranted jobs earn around 18% less per hour than those in task-warranted positions. The problem worsens for those pushed into non-graduate jobs, where they face a 31% wage penalty compared to their peers in roles that match their education. These mismatches also reflect broader issues in Singapore’s education system. OLS regression analyses of literacy and numeracy scores highlight differences in skill levels among graduates from different educational tracks. It measures how different schooling systems impact literacy and numeracy by ages 18-20. Students from certain tracks, particularly Type 2a (North American) and Type 4 (mixed) systems, score significantly lower in both areas. Type 2a graduates score 1.210 standard deviations lower in literacy and 1.074 lower in numeracy, while Type 4 graduates score 0.765 lower in literacy and 0.709 lower in numeracy compared to those from Type 1 systems. This suggests that some education tracks may not be equipping students with the skills they need to succeed in high-value jobs. 
Singapore’s system of early academic tracking—where students are sorted into different pathways based on performance—may contribute to this problem. While tracking helps train workers for specific industries, it also creates rigid divisions in career choices. The fact that students from weaker education tracks continue to perform worse in literacy and numeracy after entering the workforce suggests that these gaps are not easily closed. If these graduates are more likely to end up in task-unwarranted jobs or lower-paying roles, this may limit their long-term career mobility. Despite these concerns, the steady rise in Singapore’s graduate workforce implies long-term opportunities for economic growth. The challenge now is to ensure that higher education leads to meaningful employment. One approach is to increase career mobility by providing more opportunities for workers to shift industries through skills training. Strengthening partnerships between universities and employers (such as more "work placements", semester-internship that leads to full-time employment) could help ensure that degrees align more closely with industry needs. Singapore’s education policies have effectively grown the country’s skilled workforce, but gaps in job quality and skill acquisition remain. Without addressing these disparities, many graduates risk being trapped in jobs that do not fully utilize their potential. Policies that connect students’ majors more directly to job opportunities could help create a more efficient workforce. |
|
#Bottom-up Approach to Increasing Innovation and Competition in Innovation
In “Tapping into Talent: Coupling Education and Innovation Policies for Economic Growth” and “Measuring the Characteristics and Employment Dynamics of U.S. Inventors,” we learn that better educated individuals are more likely to be inventors. In addition, we find that many minority groups are underrepresented in innovation, namely women, black people, and poor people. By increasing accessibility of education and combating discrimination, we can drastically increase the supply of inventors. From the first paper, we model that there are talent, financial, and preference constraints related to the decision to pursue a PhD or not. Within the paper, we have a clear policy implication: since there is a group of people who want to and are talented enough to pursue PhDs while being financially prevented from doing so, we should have a policy that pays for their education, as to efficiently utilize talent. However, the paper makes the assumption that preference constraints are immutable. If we factor in our understanding that certain minority groups are underrepresented as inventors, even ceteris paribus for wealth and talent, we might attribute some of this to lack of preference. Research dating as far back as the 50s suggests that in unequal societies, members of an oppressed minority group may display internalized discrimination. This was shown in the famous “doll test”, of which the results are included in the attached table. We would imagine this internalized discrimination (i.e., thinking “people of my minority group don’t seem to succeed in research, so I shouldn’t pursue it”) would affect preference for research. If we relax the assumption of preferences for research being immutable, then we can consider increases in the “quality” of education. Since preferences are heterogeneous, we can also improve preferences for research in those more hesitant to pursue it due to discrimination. Earlier in this class, we mentioned a difficulty in patent policy: we have to grant monopoly rent as an incentive for innovation, but that will cause monopoly distortions. With the increased supply of inventors, the market for innovation would naturally become more competitive. This would “bid down” the price for innovation, and would either decrease the monopoly distortion of the patent granting exclusive rights (i.e. if we consider the length of a patent granting exclusive rights to be the price, it might decrease from 10 years to 5 years), or maybe disappear entirely (i.e., a one time flat price the government pays to make a patent public). |
|
Overcoming Innovation Barriers in Remote Collaboration Previously, we learned that technology evolves through a recursive structure, where a central backbone supports smaller subassemblies, and innovations are further built with these existing systems. For technology to evolve, the new innovation must align with existing structures rather than exist in isolation. The article “Remote Collaboration Fuses Fewer Breakthrough Ideas” this week highlights how remote teams struggle with early-stage innovation where ideas are tacit and unstructured. The challenge is not a lack of creativity, but the absence of frameworks for merging diverse knowledge into cohesive breakthroughs. Drawing from technological evolution, I propose strategies to help remote teams overcome these innovation barriers. Remote teams struggle to process tacit knowledge and unclear ideas because that often requires close back-and-forth communication and collaboration, which are naturally limited in remote settings, making it hard to streamline knowledge into clear, structured ideas. To address this, we should design ideation stimulating frameworks that help facilitate the refinement of tacit knowledge into structured ideas. Methods like Design Thinking can provide structured pathways to facilitate effective ideation, guiding teams to integrate diverse knowledge sources and define clearer ideas, helping remote teams overcome early-stage knowledge fragmentation. Furthermore, we can structure remote teams like modular technology systems, where different units operate in a layered, hierarchical framework. In this model, a central innovation hub serves as the foundation for deep ideation, while modular ideation teams work in parallel to refine specific aspects of an idea that aligns with the core vision. Each modular team is further divided into specialized units to integrate different knowledge streams, ensuring their contributions fit well into the larger framework. Additionally, workflow systems should be designed to facilitate constant, structured communication. Finally, the article highlights that less experienced researchers are often assigned technical tasks rather than being involved in idea generation. This reduces creative disruption over time, as new talent is not exposed to the critical process of conceptual innovation. Organizations should actively integrate junior members into ideation processes through junior training programs and initiatives such as "idea presenting pitch day", where newer members are given the opportunity to contribute directly to early-stage development. I made a model to show how my strategy help creative disruption: 
I = Innovation Output (number of breakthrough ideas, patents, cited, disruptive advancements). K is Harmonious Knowledge integration By using a purpose built mechanism to facilitate ideation in remote work, K increases as it helps teams refine tacit knowledge, and modular team structure improves fit of innovation with greater framework, further improves K. By including junior workers in ideation, H will improve in the long term. |
|
Tapping into Talent studies the question of how innovation and education policy affect individual career choice and aggregate productivity. The paper studies the frictions that are preventing companies from tapping into human capital; as Professor Akcigit argued in class, investing in R&D without nurturing talent is like building a car without an engine. Furthermore, Measuring the Characteristics and Employment Dynamics of US inventors, gives us a detailed demographic breakdown of inventors. We see broader social patterns occur within the inventor communities, as there are less females and ethnic minorities with patents, for example. In my discussion question from earlier this week, I asked a question about the correlation between inventor success and social status. How can we grapple with these complex and interconnected problems through policies? In this memo, I am interested in analyzing the inefficiencies when transforming human potential into human capital from a psychological perspective. In class, we have used statistical data and models to determine some levers that can impact innovation: such as parental background, education, parental income, etc… In this memo, I hope to add onto the statistical work with experiments and field work. I recently read a few chapters from Claude Steele’s Whistling Vivaldi. Chapter 1 and 2 discuss how Steele, who worked at the University of Michigan, studied the possibility of social influences on discrepancies in “intelligence” of different minority groups through multiple testing nodes: SAT scores before attending UMich versus academic performance at UMich (black students versus white students), or performance on math tests when gender stereotypes are enforced versus not enforced (female students versus male students). I want to highlight one of Steele’s recent studies on how stereotypical cues impact gender participation in computer science. 
The studies examine the role that stereotypical computer science environments play in communicating stereotypes and a sense of ambient belonging to potential majors. The results demonstrated that these kinds of environments broadcast a masculinity that made women feel like they do not belong in the field. But, when the stereotypes of computer scientists were altered through the objects in the environment, women had the motivation to engage computer science as a possible future pursuit. (as a postgraduate pathway; we learned in class that PhD’s are now the strongest mechanisms for innovation.) The results of this study make me reform my initial question in the context of this class. Which social stereotypes, if any, are restricting minority groups from becoming inventors? Can we examine the relationship between “reduced confidence” and “reduced innovation” due to stereotypes of masculinity or racial dominance, or other forms of discrimination? Is there a way to perform a psychological experiment to analyze this? Steele’s study: https://psycnet.apa.org/doiLanding?doi=10.1037%2Fa0016239 |
|
Stanford Dropouts, Age, and Creative Disruption in their Entrepreneurial Innovations Traditional policies often assume a direct pathway from higher education to technological advancement. However, the phenomenon of Stanford dropouts – entrepreneurs who leave an elite institution like this to launch transformative startups – challenges conventional knowledge about education’s role in fostering disruptive innovation development in the market. My memo will explore the implications of drop-out led entrepreneurship and age on creative, disruptive innovation development. The “Tapping into Talent” text states that education and innovation are both key to economic growth, arguing that PhDs are 20 times more likely to become inventors compared to the general population. If someone earns a PhD is determined on multiple factors, including one’s IQ and parental income, as stated by the text. Along with this, the “Nostalgia Effect” text argues that disruptive creativity declines with academic age – suggesting that younger individuals are more likely to produce paradigm-shifting research because of their openness to challenging existing ideas. Stanford has become synonymous with Silicon Valley entrepreneurship, producing a long list of founders who left school to create industry-defining companies. These individuals are often young, willing to challenge existing ideas, and these cases suggest that, in certain high-tech sectors, the most innovative individuals may not benefit from completing traditional academic programs like a PhD. Instead, they find success by leveraging early-stage ideas and small, disruptive teams. This trend seems to be paradoxical when contrasted with research illustrating that PhDs are 20 times more likely to become inventors than the general population. Similarly, policies aimed at expanding PhD slots, while fostering new researchers, may dilute the average talent level, leading to lower productivity growth. The Stanford dropout model challenges this notion, demonstrating that age and environment, rather than formal educational attainment, may be key drivers of disruptive innovation – which is supported by the “Nostalgia Effect” text. As seen by the chart below, at Stanford, most of their alums pursue entrepreneurial endeavors at the highest frequency right after they leave Stanford, with it slowly decreasing as they get older. 
I argue that R&D subsidies play a key role in generating long-term innovation capacity – an idea slightly different than that argued by the “Tapping into Talent” text, which argues that R&D subsidies provide immediate, but limited effects. For instance, these subsidies can help form elite universities like Stanford as talent incubators. Rather than serving solely as training grounds for PhDs, institutions like Stanford function as network hubs, allowing future founders to form teams, access VC opportunities, and gain credibility. Along with this, while PhDs engage in deep research, dropouts often prioritize speed-to-market and disruptive problem-solving approaches, which aligns with findings from the “Large Teams Develop and Small Teams Disrupt Science and Technology” text that small teams are more likely to introduce paradigm-existing ideas, while large teams leverage existing knowledge. In the chart and table below, we see that, at Stanford, founders with less educational credentials tend to raise the most funds, have the highest IPO and M&A valuations, highest market caps, and end up employing the most people in the long-run (think Google and how they have now, eventually, adopted a hierarchical team structure and employ thousands of individuals). 
Additionally, the concept of disruption (D) can be quantified as shown below, where D measures the extent to which new ideas replace older ones. Small firms exhibit higher D values, reflecting their ability to challenge existing paradigms. In contrast, large firms tend to consolidate knowledge, resulting in lower D scores. This emphasizes the importance of creating small, flat teams to sustain disruptive innovation. 
These flat teams, where leadership roles are distributed, produce more novel and disruptive scientific contributions compared to hierarchical teams that are often seen at larger companies, as argued in the “Flat Teams Drive Scientific Innovation” text. This aligns with how successful Stanford dropouts often build teams that challenge incumbents, such as OpenAI being started by a Stanford dropout that challenges larger companies like Google (another startup created by a Stanford dropout). As startups like Google mature into larger corporations, they tend to transition into hierarchical structures, focusing more on incremental rather than disruptive innovation. Stanford dropouts produce some of the highest levels of breakthrough innovations, as “no U.S. school produces more founders of billion-dollar firms” (Alexander, 2019). What we can learn from this, and what I argue, is that R&D subsidies should be used to nurture entrepreneurial ecosystems, in conjunction with emphasizing PhD production – not simply having a heavier focus on PhD programs. |
|
Koning et al. show that when women are underrepresented in invention, society misses out on innovations that address women’s needs. Their analysis of U.S. biomedical patents reveals that female researchers—especially in all-female teams—are more likely to develop patents focusing on women’s health. These findings hint that the question “Who gets to invent?” also shapes “Who benefits from innovation?” This observation is highly relevant for the field of artificial intelligence (AI). In AI, men still dominate the research and development workforce. Thus, there are similar concerns that the relative lack of women in AI might overlook the distinct needs of female users or communities. I plan to examine this issue in my final project by analyzing AI startups with varying proportions of female founders. The goal is to see whether higher female participation is linked to technologies that address specialized societal problems—such as maternal health, women’s financial inclusion, or even female-centric safety apps. Although existing research on AI fairness often highlights data bias, focusing on invention teams could offer a complementary explanation for disparities. If more women lead AI ventures, we might see not just fewer biases but also more solutions tailored to female perspectives. As a small demonstration, I used a simulated dataset to show how differences in female team representation can predict the proportion of solutions aimed at women’s issues. The short R script below simulates a basic linear regression. It models a hypothetical “female-focus score” in AI solutions as a function of female founder share, controlling for overall R&D budget. While only illustrative, this approach sets the stage for deeper empirical inquiry using real data on AI firms and their product lines. Next steps include gathering pitch descriptions, patent data, and user surveys to measure the actual impact of female-driven AI ventures. Ultimately, insights from this work might inform policymakers and investors who wish to foster inclusive innovation ecosystems—ones that reflect the diverse experiences of all potential inventors and beneficiaries.
|
|
Firm Size, Productivity, and Turnover: A Comparative Analysis of Germany, Italy, Norway, and Türkiye In “Large Teams Develop and Small Teams Disrupt Science and Technology,” the researchers illustrate how team size influences the nature of innovation, with small teams driving disruptive breakthroughs while large teams focus on incremental advancements. The paper also delineates the trend where teams are growing larger and larger in all areas while smaller teams and individuals are diminishing, calling attention and effort to sustaining the heterogeneity of team sizes. The disparity between large and small teams in their approaches to innovation also mirrors dynamics in the broader economy. In Professor Akcigit’s lecture, we see how incumbents and new entrants differ in their innovativeness and productivity, with incumbents often using their market power to stifle competition by eliminating more productive entrants. This interplay between firm size, innovation, and competition forms the foundation of my inquiry. In this memo, my hypotheses are 1) Compared to larger firms (new entrants) the smaller firms (new entrants) will not only have a higher turnover rate but also higher labor productivity due to their inventiveness. 2) Different countries will differ in their dynamics due to the national policies, institutions, economic progress, the nature of incumbents in the country, etc. While Professor Akcigit complicates the creative destruction model by emphasizing that a firm’s age—rather than just its size—is a key factor, my analysis focuses on firm size due to data availability. What sets this analysis apart from our previous readings and coursework is my comparative approach: I examine firm dynamics across four countries, placing their statistics side by side. This cross-national perspective allows me to assess whether patterns related to firm size and economic dynamism are generalizable across different institutional and cultural contexts. The results reveal that small firms, particularly those with 1 to 19 employees, are the most dynamic. They constitute the majority of firms in all countries and exhibit the highest turnover rates. However, when it comes to labor productivity, a surprising trend emerges: across all countries, larger firms tend to be more productive, with labor productivity steadily increasing as firm size grows. Another noteworthy observation is that while countries display similar internal patterns, stark differences emerge when comparing them against one another. For instance, Norway exhibits significantly higher labor productivity than the other countries in the sample, regardless of firm size. Given that Norway also has relatively fewer firms, it would be valuable to explore the structural, institutional, or policy factors that contribute to its impressive productivity. These findings invite several avenues for further research. First, the relationship between firm size and labor productivity raises questions about the underlying mechanisms driving this pattern—do larger firms benefit primarily from economies of scale, better resource allocation, or superior access to technology and skilled labor? Why didn’t the volatility of small firms bring them higher productivity, as expected? Second, the case of Norway suggests that national policies, labor market structures, or industrial strategies could play a pivotal role in shaping productivity outcomes. Investigating these factors could offer policy insights for other nations seeking to enhance labor productivity while maintaining a dynamic entrepreneurial ecosystem. 
|
|
The papers discussed this week provided models supporting the idea that older researchers tend to slow the “growth rate of ideas” in various fields of research. When new researchers enter the field, they take on ideas from recent papers, then cling to those ideas over the span of their research careers, which is generally their entire lifetimes. This results in older research papers and older researchers dragging back the rate of growth of their fields. 
We can model the step size, s_t, as a function of the average age of cited research. s_0 is some base level of research. Gamma and delta are some constants that tell us the magnitude of the effect and the sensitivity of research age on the step size. A_max is the 10-year allowed maximum paper age per the policy, and A_ is the current average age of papers that are used as research papers, generally older than 10 years. By this model, restricting research age results in an increase in the step size of innovation, which can be interpreted as an increase in the novelty and impact on the field that new papers generate. 
This formula tells us the effect of this policy on the growth rate of the field. Lambda, the innovation rate, should stay roughly the same. Then, the growth rate of the research field, g, increases as s_t increases. The policy and model are preliminary. They require the assumptions that the innovation rate remains the same, that the age of research cited is strongly correlated to the step size of innovation, and that restricting the pool of citeable research only affects novelty and has no negative effect on the quality of research or rate of innovation. These are all very strong assumptions. However, the model does align with the growth rate of the field of AI research. After Hinton et al.'s 2012 breakthrough in neural networks, most AI papers began citing more recent research that built upon ideas introduced since that paper, which itself has over 100,000 citations. At the same time, the AI field has arguably been the fastest-growing area of research: step size is higher, resulting from a focus on work made more recently than in 2012. |
|
While we've read that the consolidation of capital among large incumbent firms stifles innovation through non-productive strategies, this theory may have different implications for different industries. Firm size and technological innovativeness are broadly negatively correlated; however, that is particularly true for nascent industries, and more complicated in traditional industries like basic manufacturing. Smaller firms dominate in disruptive innovations, but primarily in tech-oriented, low capital intensity industries. Large incumbent tech firms seemingly misallocating talent or underinvesting in R&D still benefit through investment or acquisition of these "unicorns." Earlier readings would suggest that large incumbent tech firms stifle innovation by underinvesting in R&D and not hiring the right technically gifted inventors or product designers; however, large tech firms invest significantly in R&D and the proportion of unicorns across the world hailing from tech origins is overwhelming. If these firms spend billions on R&D and hiring the best inventors, and are still outdone by a smaller firm that is better suited to generate disruptive new ideas, the capital accumulation and dissemination to and from these large tech firms to smaller successful innovators should be encouraged. The low capital intensity of generating disruptive new innovations today makes innovation more accessible, meaning access to capital will not generate a moat for large firms in tech. Perhaps we should adjust our model for thinking about the interplay between firm size and innovative capacity, and consider whether or not large firms stifling innovation with non-productive strategies can be generalized across all industries. The tech industry is an interesting case for this, where venture capitalists, incubators, and large tech firms all contribute capital to disruptive and innovative new startups. The generally low capital intensity of this industry also incentivizes inventors to develop products with minimal external capital resources, allowing them to own the profits of their invention more so than in other capital intensive industries like manufacturing or consumer retail, attracting more talented inventors to the industry. Additional considerations like cross-pollination between industries, with tech applications in many other industries, can help explain why we see so much successful and disruptive innovation in the tech industries over other more capital intensive industries, and why large tech firms are incentivized to acquire young startups as they experience success rather than to over develop on human capital, which could be seen as a sort of misallocation of talent. 
References: https://iap.unido.org/articles/firm-size-technology-and-trade-policy |
Impact of Gender Equity Policies on Hiring Practices and Patenting ActivityThe characteristics of individual inventors we discussed in Tuesday’s lecture seemed pretty intuitive — high education, affluent parental background, and high IQ and talent. I found the distinction between substantive and transformative entrepreneurs an important addition to our conversation about how to empirically measure innovation. Entrepreneurship can capture the added business value of innovation that may not be formalized in patenting activity. Differentiating the data (by firms that hire at least 1 R&D worker and firms that don’t) controls for non-innovative entrepreneurship. Moreover, I liked that Thursday’s readings expanded upon the question of “who becomes an inventor” to explore characteristics of teams that are associated with innovation. Connecting across the readings reveal potential compounding effects between remote environment and nostalgia bias such that reduced spontaneous interaction and reinforcing existing paradigms discourages disruptive innovation. The paper “Who do we invent for? Patents by women focus more on women’s health, but few women get to invent” stood out to me. The discussion of gender gaps as a cultural dimension impacting innovation adds nuance to findings about inventor demographics. Since “a lack of representation among inventors translates into a lack of breadth in inventions,” it is beneficial to overall societal innovation to increase representation among inventors. However, there are systemic barriers to access that limit educational and professional opportunities for women. I am interested in exploring statistical associations between gender inequalities in hiring practices and the gender gap in patenting activity across industries. Specifically, I am curious if there are noticeable differences before and after gender equity policies are enacted for hiring and patenting trends. Some notable gender equity policies include Title IX and its expansion to STEM education and workforce participation and the America COMPETES Act (2007, reauthorized 2010). I propose the following generalizable regression which can examine pre- and post- impacts of specific policies: 
|
|
The article “Large teams develop, and small teams disrupt science and technology” by Lingfei Wu, Dashun Wang, and James A. Evans introduces the idea that small teams are more disruptive, while larger teams often focus on refining or extending current knowledge. This can be explained by the different incentives each group has when innovating. Large groups often have large funding and higher pressure for results, as investors expect positive returns on their investments. This means that the focus shifts from genuine discovery to a search for validation to secure more funding and citations for the researchers. Since many people are searching for results and there is a push to produce confirmatory findings quickly, we risk “p-hacking”, the selective reporting of analyses or hypotheses until a statistically significant result emerges. 
Another angle we can consider involves the concept of Bayesian Updating. Large teams face pressure for quick results and must therefore focus on the short term. This may hinder the possibility of long-term discoveries. This can be expressed as an equation: 
Where: Large teams have very strong priors, as they typically investigate well-established research areas with many earlier discoveries. This makes sense, as investors are more likely to commit to ideas that have worked before. Since their priors are strong, a significant amount of evidence is required to change their beliefs, making them more likely to achieve only marginal updates to existing knowledge. Small teams, on the other hand, have almost no priors because, to compete with large teams, they must explore new fields or be highly disruptive. If they make a radical new discovery, they are more willing to pursue it, a§s there is less consolidated work on the topic. A historical example is the rise of Impressionist art. New artists such as Monet and Degas were highly disruptive and often clashed with established art institutions. These institutions had funding and were adamant about continuing with traditional art. They were less willing to commit to a new style that their benefactors might dislike and feared that embracing change would cause the organization to lose prestige. In the context of creative destruction and Schumpeterian economics, one realizes that if large companies are allowed to amass too much power and the forces of preservation become too dominant, disruptive ideas will be suppressed. For instance, in the Impressionist art movement, the French government offered sponsorships and subsidies to large institutions, thereby suppressing many potential artists. This highlights the importance of balancing the forces of preservation, creation, and destruction to ensure both disruptive and incremental developments in the economy. |
|
The paper “Measuring the Characteristics and Employment Dynamics of U.S Inventors” underscores interesting observations in the demographics of inventors, their employers, and their earnings. 
From this figure it is immediately apparent that employees with a Democrat political affiliation are much more prominent in the tech field than those with Republican affiliations. There is also a much higher Democrat presence in the tech sector than in Wall Street banks (which are roughly 50-50) and large enterprise companies; these other sectors were not included in the figure above. |
|
AI Patent Growth and R&D Intensity Across High-Tech Industries These two graphs illustrate trends that connect directly to the paper’s findings about where inventors work, how they are compensated, and how innovation is distributed across different industries. The first chart shows the dramatic increase in annual granted patents related to artificial intelligence (AI) since 2013. The stacked formatting displays separate industries, with personal computing devices comprising the largest segment, followed by telecommunications, life sciences, transportation, and other areas. The paper emphasizes that inventors are increasingly found at older, larger, and more established firms. The rapid rise in AI patents aligns with the observation that big companies with substantial resources tend to dominate high‐tech research. It also echoes the paper’s finding that younger inventors are less likely to be at start‐ups, which suggests that many of these newly granted AI patents could be coming from incumbent tech giants rather than small newcomers.
Furthermore, the first chart’s steady upward slope demonstrates an overall growth in AI‐related patenting that mirrors broader shifts in the global economy toward data‐driven technologies. The paper shows that a significant portion of inventors, especially super star inventors, receive high compensation when working at well‐established firms. The exponential growth in AI patents may be a result of these larger firms drawing top talent with superior pay and better R&D facilities. This feeds back into the paper’s narrative that there is a rising concentration of inventive activity at the upper tail of the earnings distribution. 
The second chart, which compares capitalized R&D as a percentage of invested capital across various sectors, helps explain why certain industries attract more inventor talent and patents. Pharmaceuticals and biotechnology top the list with over 100 percent, meaning they invest more in R&D than their total invested capital. Internet software, system and application software, semiconductor equipment, and retail follow in descending order. High R&D intensities in biotech and software are consistent with the paper’s findings that inventors tend to be compensated well in R&D‐focused industries. The paper notes that inventor pay is highly skewed, with a large fraction in the top 10 percent or top 1 percent of earnings. This lines up with the fact that biotech and advanced software companies rely heavily on breakthroughs and proprietary knowledge, fueling demand for top inventors. Larger incumbents in these industries are willing and able to compensate talent generously. This is exactly what the paper observes when it describes the strong correlation between inventive productivity and earnings, as well as the growing share of patents tied to these bigger market players. In short, the two graphs reinforce the paper’s argument that innovation is concentrating in specific high‐tech and R&D‐intensive industries, often inside larger firms that offer top‐tier wages. The surge in AI patent grants and the outsized R&D spending in biotech and software are prime examples of how inventive activity is clustering where investment is highest and where top inventors can capture significant returns. |
A Cross-country and Over-time Analysis of Business Financing Access of Men and Women EntrepreneursThis week we discussed in class factors affecting who becomes an inventor and entrepreneur including education, family income, etc. My hypothesis is that access to funding is also an important criteria for entrepreneurship, as startups or innovation groups that lack funding often struggle with hiring talent, R&D, and growing the business. Last week I analyzed a dataset that indicates earlier access to VC funding is a key indicator for startup success. And this week, I found another dataset on OECD website about business financing for people who start a business. The dataset spans over decades and includes information about the percentage of financial account holders by country by gender among entrepreneurs who’re over age 15; and the percentage of borrowing needed by country by gender among the same group of entrepreneurs. I cleaned the dataset and investigated 2 main questions. First, how does financial account holder percentage differs by gender in the U.S. across 10 years. Second, is there any gender gap between access to financing (as indicated by percentage of financial account holders, or alternatively, percentage of borrowing needed among entrepreneurs in all countries in a specific year.) The former percentage of financial account holders should be positively correlated with access to financing, while the latter factor borrowing needed should be negatively correlated with access to financing, because debt is often used when other forms of funding is unavailable for entrepreneurs. The 3 charts I graphed are attached below. 
From the year of 2011 to the year of 2021, U.S. women entrepreneurs’ access to financial accounts have been increasing in general, with small decrease between 2014 and 2017. Men’s access has been relatively stable over the years, initially much higher than women until 2013, but then quickly surpassed by women since 2018 onward. U.S.’s data seems to suggest that the gender gap between men and women entrepreneurs’ access to financial accounts have been closing. But this may not tell the story across the globe, as U.S. is usually considered to have higher gender equity in the labor market. 
The second chart I generated indicates financial account holder percentage by country and gender in the year of 2017. We can see that in most countries, men entrepreneurs have a higher percentage of holding financial accounts compared with women (most blue bars exceed pink bars in length). Western and Northern European countries, North American, and developed Asian economies have higher women financial account accessibility than Eastern European, South American, and Middle Eastern countries. The most prominent gender gap is in Turkey where only have of women entrepreneurs have financial accounts, compared with more than 80 percent for men. Mexico, Costa Rica, and Colombia also demonstrate large gender gaps while having overall lowest accessibility among both men and women entrepreneurs. 
The third chart demonstrates borrowing needed by entrepreneurs in different country in 2017. We can see that a higher percentage of men need borrowing than women in most of the countries. Interestingly, this time, European countries tend to have higher demand for borrowing compared with both North American countries and South American countries, which might arise due to specific borrowing environments in those countries. Gender gap in this category is most prominent in Spain, Italy, Korea, and Slovak Republic. I concluded that gender does have an impact on entrepreneurs' access to financing. Next week, I would like to explore how business financing exactly affect entrepreneurship activities in different countries. I plan to build financial access indexes for both genders in each country based on the data above, and then compute gender gaps data. I will collect additional dataset as indicator for entrepreneurship and innovation level for each country, and tie it back with the financial access index and gender gap to explore any correlations. Finally, I hope to add other variables such as education, government policy, and culture to build an entrepreneurship index for each country. |
|
In this week's reading, "Flat teams drive scientific innovation", we learn about how the organizational structure of research teams has an impact on their effectiveness in creating significant, novel patents. Through analyzing the L-ratio of various teams, which represents the ratio of members playing leadership roles to total team size, they found that flatter, egalitarian teams provided more novelty and long-lasting innovations. However, on the other hand, tall, hierarchical teams more often developed existing ideas and benefited the researchers with the highest authority in papers. |
|
Business vs. Government R&D “Large Teams Develop and Small Teams Disrupt Science and Technology” and “Flat Teams Drive Scientific Innovation” clearly articulate how differences in research environments shape innovative activity. In particular, the former emphasizes the importance of small research teams that are more likely to search deeply into the past and produce disruptive works, while the latter emphasizes the importance of flat, egalitarian teams that output more novel findings. Given the broad concerns over scientific stagnation in the 21st century – growth in publications is accelerating, but new ideas are lacking – these findings warrant an investigation into the factors contributing to larger and taller R&D teams. One dynamic I want to discuss and explore is the shift in the U.S. between federally funded R&D and business-funded R&D. Viewing data I pulled from the National Center for Science and Engineering Statistics, it’s clear that R&D intensity as a proportion of the U.S. economy has been on the rise over the past 40 years. Crucially, however, federally funded R&D has shrunk to 18% of total R&D expenditures in 2022, down from its heyday of 66.5% in 1964. The dominant driver of R&D intensity has been the business sector, which rose from 30.8% to 76% over the same period. 1964 was the golden age for government-sponsored research as the National Laboratories and the Defense Advanced Research Projects Agency, among others, pushed the frontiers of American research during the tense Cold War era. As the United States government scaled back on this funding, the 1970s also saw the emergence of venture capital models and later large-scale technology companies with quasi-monopoly profits that they could funnel into experimental labs. On the one hand, we would expect business R&D to feed into more applied research environments, likely larger and more hierarchical. Indeed, the “Large Teams” article posits that “[large business organizations] may focus on sure bets with large potential markets.” That being said, the allure of a robust venture financing ecosystem has fielded significantly more high-velocity startups that are by nature smaller and flatter. Business R&D may not be inherently worse for innovation, but incentives for truly novel ideas to emerge may indeed be muted by a shareholder-sensitive focus on bottom-line performance. Separately, we may be entering yet another age of ramping federal R&D. During the Biden administration, there was a concerted effort via the CHIPS and Science Act to address geopolitical shifts and competition abroad; to note, there was a 13% jump in federal R&D spend from 2022 to 2023 not captured in the graph. Certainly, there seem to be possible optimizations at the environment level for U.S. innovators. Further investigation should be conducted on the incentive and organizational structures for businesses in particular given their outsize role in capital allocation. 
|
Innovation in Flat vs. Hierarchical Organizations: Google vs. IBMThe article "Flat Teams Drive Scientific Innovation" by Xu, Wu, and Evans examines the impact of team hierarchy on scientific innovation. Through the L ratio, which measures the proportion of team members in leadership roles, the authors find that flatter, egalitarian teams produce more novel and disruptive scientific contributions. On the other hand, hierarchical teams (low L ratios), tend to develop existing ideas, boosting lead scientist productivity at the cost of more junior level researchers. This week, I apply some of these ideas from this paper to examine innovation across two companies: Google and IBM. Google has touted flat organizational structures as a key part of being “googly” and has reduced manager roles over time. IBM, on the other hand, is more traditional, with higher proportions of middle managers. I compare two metrics to measure innovation.
Google, the flatter company, has 1.64 times more patents per employee compared to IBM, the more hierarchical organization. This finding aligns with Xu’s, Wu’s, and Evans’ paper, which suggests that flatter teams drive more innovation–at least when using patents as a measure of innovation. That is, a flatter structure enables individuals to contribute more directly to innovative efforts. On the other hand, a more rigid hierarchy might lead to bottlenecks in decision-making and innovation, where only some individuals drive patent activity (instead of a larger population of the workforce). For the chart on the right, Google’s higher average citation counts suggests that their AI research is more influential and disruptive to the literature than IBM’s. This further supports the flat vs. hierarchical argument from the paper, where flatter teams generally contribute more to novel, innovative research. The article also argues that flat teams’ papers have longer term impact, and we could potentially use citations as a proxy for that. The sample size of seven is too small to conclude much about the graphic on the right, so this is certainly something that I would explore more. Also, correlation is not causation, so ideally we would isolate the effects of flatness, such as through an L ratio. In the corporate world, the arguments made about flat and hierarchical teams seem to hold up, with Google scoring higher than IBM on innovation, measured through patents per employee and average citations. More rigorous data collection would enable us to nail down these conclusions. Here is the link to my Google Sheets. |
|
Balancing Disruption and Development: Innovation Dynamics in Government Institutions The paper "Large teams develop and small teams disrupt science and technology" explores how larger teams excel in developing existing knowledge, while smaller teams are more likely to disrupt by exploring new problems and opportunities. I began to wonder about these dynamics beyond the private sector and the implications for government size. Complaints about the evergrowing reach of the United States Government can be frequently heard among top business tycoons, as the US government employs approximately 15% of the U.S. workforce, raising important questions about innovation in public institutions, while private sector employment stagnates. 
The first chart, which tracks R&D spending per employee from 1976 to 2023, reveals that agencies like NSF and NIH consistently allocate significantly higher amounts to research and innovation per employee. This reflects their research-intensive mandates and specialized focus on advancing science and technology. On the other hand, agencies like USDA and DOT exhibit lower R&D spending per employee, aligning with their broader operational goals and administrative responsibilities. |
|
Equality in the workplace: Stemming the leakage between PhD attainment and innovation In “Measuring the Characteristics and Employment Dynamics of the U.S. Inventors”, we are introduced to the demographic makeup of inventors. I was particularly interested in Figure 1 on inventor female share. It was rather shocking that the share was still disproportionately low, at only around 16% in 2016. Whilst it’s growing, it’s still a worrying statistic. Particularly, if we were to refer to the model introduced in the first paper, “Tapping into Talent”, where gender is not factored into the equation, it made me question how gender dynamics play into the transmission mechanism between getting a PhD and hence innovation. Notably, the model seems to lump together males and females into a homogenous representative agent that becomes a team leader at λ and hence generates ideas at a rate of: q(z, t) = φz^η1 * n^η2 * a^η3 I hypothesize that in fact, there is a unique “leakage” between women getting PhDs and thereafter, idea production. This would have implications on policy, as it presents another cost effective avenue through which the government can attempt to induce idea production. Namely, by preventing this leakage. Below is a graph comparing the amount of doctorates earned by women vs men in specifically Science and Engineering Fields. I had originally found the ratio for all PhDs, but found that most of these innovations came from those with S&E backgrounds, which made sense as a math/CS PhD is perhaps more applicable to AI/ML breakthroughs than a degree in english. Hence, to lessen the noise, I specifically found data for S&E fields.
Figure 1: Ratio of women to men receiving doctoral degrees in S&E fields This result was rather confounding, as it showed that the share of female S&E doctoral degree holders was way larger than the inventor female share. The proportion of female S&E doctorate holders increased from 38% in 2001 to around 42% in 2010 and maintained thereafter. This pales in comparison to the 16% share of female inventors in 2016. It hence becomes clear that despite some positive trends in women attaining advanced degrees, this does not commensurately translate into more women inventors. Of course, there could be many reasons for this, such as unequal wage deterring women from taking on often exorbitant patent filing costs, or bias in the patent application process. (Here’s a post which goes into greater detail: https://inventiv.org/blog/inequality-disparity-in-patent-applications-issuance-women-inventors) Clearly, the transmission mechanism between attaining a PhD and becoming a “team leader” and idea generator does not hold in reality. This is not to discredit the first study, but perhaps open a new avenue the government can take advantage of to increase idea generation. While increasing the access people have to attaining a PhD, the government should also look at ways to alleviate the insidious barriers that prevent PhD holders from generating innovative ideas. From “softer” policies that ensure an unbiased patent review process, to ensuring equal wages. In addition to increasing the absolute number of high skilled workers, we should also focus on strengthening the ability for all PhD holders to become innovators. Going back to our original equation: q = φz^η1 * n^η2 * a^η3 This would amount to increasing the poisson rate λ at which women become team leaders, which would increase z. n would also increase as they would be more incentivized to stay on in research roles. Φ might also increase due to having a more diverse team. In conclusion, concurrent with policies that increase the amount of skilled labor becoming future innovators, we should also focus on policies that directly target the transmission mechanism that creates innovation from this skilled labor. It is not enough to generally increase the number of PhD holders, if it does not lead to a commensurate increase in innovation as some groups are systematically disadvantaged from doing so. Conversely, it is only by pushing for an equitable environment that we can encourage innovation. |
|
AI, Innovation, and Family-Owned Businesses in Japan Recent literature on economic growth and innovation policy has highlighted the importance of human capital, team structure, and gender disparities in technological development. The article Tapping into Talent (Akcigit et al., 2024) emphasizes the synergy between education and innovation policy, while Wu et al. (2019) and Xu et al. (2022) explore the role of team composition in scientific disruption. This memo applies these insights to family-owned businesses in Japan, examining how AI adoption, team hierarchy, and R&D investments impact innovation and economic sustainability. Japan’s R&D Investment and AI Adoption Japan consistently ranks among the top five countries in R&D and higher education spending, investing approximately 3.84 trillion yen in 2022. This substantial investment has propelled Japan as a global leader in technological innovation. However, the benefits of AI adoption and technological advancements are not evenly distributed. While large corporations integrate AI seamlessly into operations, smaller family-owned businesses—comprising over 90% of Japanese companies—face barriers to adoption due to resistance to change in traditional and hierarchical structures that may be hindering access/limitations to AI adoption. While this stability in family structured businesses fosters long-term business survival, it may also hinder disruptive innovation. Akcigit and Goldschlag (forthcoming) demonstrate that inventor employment dynamics shape technological diffusion. If AI development remains concentrated within large firms, smaller family-run enterprises will struggle to compete, deepening the technological divide. Policies promoting AI training, subsidizing digital transformation, and democratizing access to AI tools for family businesses could mitigate this gap. Team Structure and Innovation Constraints Research by Wu et al. (2019) and Xu et al. (2022) suggests that small and flat teams drive disruptive innovation. However, family businesses in Japan traditionally adhere to hierarchical structures emphasizing family seniority, which may limit leadership diversity and stifle innovation. Japan’s family-owned businesses tend to have lower L-ratios (the ratio of leaders to total team members), reducing the number of decision-makers who can drive disruptive ideas. The hierarchical structure may create rigid organizational dynamics that prevent new perspectives from emerging, further limiting innovation. Conversely, research by Lin et al. (2023) indicates that in-person mentorship fosters disruptive breakthroughs, a key characteristic of Japanese family businesses. This suggests that while strict hierarchy may be a limitation, smaller team sizes and traditional mentorship models could counterbalance some of these effects, promoting incremental innovation. To analyze the impact of AI adoption and team structure on innovation in family-owned businesses, a model similar to this could be analyzed: where: I = Innovation output (measured by patents, new product launches, or technology integration) A = AI adoption level (extent to which AI is integrated into business operations) L = Leadership ratio (number of leaders to total employees, representing team hierarchy) R = R&D investment (as a proportion of revenue) T = Team structure flexibility (scale of hierarchical rigidity) A × L = Interaction effect between AI adoption and leadership diversity A × T = Interaction effect between AI adoption and team flexibility L × T = Interaction effect between leadership and team structure flexibility A × L × T = Combined interaction effect capturing the complexity of AI, leadership, and hierarchy on innovation ε = Error term accounting for unobserved influences This model account for interaction term effects that could influence innovation output, providing a more nuanced understanding of how AI adoption interacts with hierarchical constraints in family businesses. |
Team Innovation: Structuring the Remote Teams for Better InnovationLack of tacit knowledge exchange and fluid discussions 
To this, I propose a hybrid model: |
The U.S.’s OpenAI vs. China’s BaiduIn Flat Teams Drive Scientific Innovation (Xu et al., 2022), the authors argue that flat research teams foster more novel, long-term scientific impact, while hierarchical structures prioritize efficiency and incremental improvements. I want to examine OpenAI and Baidu, two leading AI firms, to see how their contrasting team structures have influenced innovation. Flat teams distribute leadership and encourage risk-taking, leading to disruptive breakthroughs. OpenAI initially operated as a flatter research lab, especially before commercialization, producing major advances like GPT-2 and GPT-3. In contrast, Baidu’s AI research is more centralized and influenced by Chinese government policy and funding, optimizing its models for state and industry applications rather than academic breakthroughs. While Ernie 4.0 is competitive, its global innovation influence lags behind OpenAI. Baidu has a higher number of AI research publications, yet OpenAI’s research has a greater impact per paper, as shown by citation data from 2010–2023:
Even with fewer publications, OpenAI’s research is cited more often, suggesting that a flat team structure leads to more influential AI breakthroughs. Given recent advancements in China’s AI sector, like DeepSeek, I want to explore whether this pattern still holds. If hierarchical structures slow long-term AI progress, flattening research teams may be key to accelerating innovation. This comparison highlights how research culture shapes AI development, with implications beyond corporate strategy to national AI leadership. |
|
Memo: Disruption and Development in Scientific Innovation Innovation is a balancing act between building on what we know and pushing the boundaries of what’s possible. In their 2019 study, Wu, Wang, and Evans explore how large and small research teams play distinct roles in shaping science and technology. Their findings reveal a fascinating contrast: large teams refine and expand existing ideas, while small teams shake things up with disruptive breakthroughs. This insight is highly relevant not just for academic research but also for industries like AI, where the composition of research teams influences the speed and direction of technological change. AI Research: A Case Study in Team DynamicsArtificial intelligence is a prime example of how team size impacts innovation. Heavyweights like OpenAI, Google DeepMind, and Meta AI drive AI development, refining models and scaling up architectures (think GPT-4 and AlphaFold). Meanwhile, smaller academic groups and startups take risks, exploring unconventional ideas like neurosymbolic AI or alternative training methods that challenge the deep-learning status quo. The contrast highlights the broader trend identified by Wu et al.: big teams consolidate knowledge, while small teams create the sparks that lead to paradigm shifts. A Simple Model of DisruptionTo capture this relationship, we can think of an equation that links team size and disruptive potential: (D = \frac{\alpha}{T^\beta}) where:
Studies suggest that (\beta > 1), meaning that disruption drops off fast when teams get bigger. This supports the idea that smaller teams are better at shaking things up, while larger teams tend to refine existing knowledge. What This Means for Research and InnovationSo, what can we take away from all this? If policymakers and R&D leaders focus too much on large-scale collaborations, they might unintentionally limit disruptive breakthroughs. A smarter approach would be to fund and support both types of teams—investing in big teams for development while ensuring that small, agile groups have the resources to experiment and challenge norms. Wu et al.’s work, combined with real-world cases in AI research, makes it clear: innovation thrives when there’s a healthy mix of builders and disruptors. |
|
Several of the papers this week investigate the relation between team size and innovation outcomes. Notably, "Measuring the Characteristics and Employment Dynamics of U.S. Inventors" reveals that inventors in mid-size young firms generate the most impactful patents and “Large teams develop and small teams disrupt science and technology” finds that smaller scientific teams create more disruptive research. Given these discoveries, I was interested to see how these trends have played out in the AI industry. To give a snapshot of the market, I used huggingface’s Chatbot Arena leaderboard to approximate current model performance. Defining the market as the creators of the top fifteen models (from distinct organizations), I estimated the number of AI-focused employees and the age of each firm. The graph gives an interesting look at the current LLM race; the AI ecosystem can be divided into three main categories:
Excluding market leaders (Google and OpenAI), there is no observable correlation between firm size and model performance. In addition, Deepseek provides a great example of a mid-size young firm that creates innovation. This graph is merely a snapshot of a constantly evolving race but gives a good idea of how young firms have been able to compete despite resource disadvantages. The existence of open source models and strengthening international competition gives hope that young innovators will continue to enter the market in a pro-competitive environment that pushes progress.
|
Considerations on Microfinance: Middle Income Financing for Middle Income CountriesIn "Who do we Invent for?", Koning et al. found that female researchers were signficantly more prolifc than men at innovating solutions for women's health. Thus, they conclude that the beneficiaries of innovation are heavily dependent on the identity of the inventor. Applying this insight to our discussions on growth economics, this memo discusses microfinance as a potential way to bridge the innovation gap in middle-income countries. Microfinance, a term often attributed to nobel laureate Muhammad Yunus (also the chief advisor of Bangladesh), is viewed is a unique solution for entreprenuers in the emerging world developed and managed by local institutions. In Poor Economics, by development economists Banerjee and Duflo, microfinance is primarily discussed as a mechanism for escaping the "poverty trap". As entreprenuers in developing countries lack access to the cheap capital afforded to those in developed countries despite having relatively small capital requirements to achieve major returns, microcredit offered a mutually beneficial solution for lenders and creditors sitting on excess cash (with little opportunties to invest elsewhere). However, recent research questions microcredits ability to generate long-term economic benefits. Microcredit has resulted in major increases in per capita debt in many developing economies without a proportional increase in productivity. These findings track insights covered in class - most microcredit financing is used on operating costs for small everyday businesses, which lack the capability to generate the disruptive innovation needed for Schumpterian growth theory. Two trends within microfinance could contribute to beneficial shifts, especially in middle income countries:
Even though total microcredit borrower growth has stagnated from peaks in the early 21st century, total borrowing volume continues to increase due to changes within the borrowing demographic. As seen in the graphic above (compiled using data from the IMF) the proportion of non-low income borrowers consitutes a growing proportion of total borrowers. This signals that microcredit is increasingly being used for non-subsistence ventures, which is critical for both infusion and innovation. Furthermore, this signals that the demand for microcredit is shifting toward middle income economies and that the former low income incomes which benefited from microcredit have entered middle income status. Due to inertia, it is likely that these institutions remain relevant as incomes continue to increase, which should lead to a continuation of this trend.
Although microcredit initially centered around small sized, short-term loans to small businesses, it has since expanded to new financing mechanisms, many of which are suitable for middle income countries. For example, data from the NBER demonstrates that the use of"digital collateral" strongly decreased deliquency rates compared to unsecured loans.
As a result, lenders had a much higher rate of return, enabling them to relax credit constraints and offer additional loan types and sizes at no cost to recipient household balance sheets. As many lower income countries lack the digital infrastructure to take full advantage of these loans, this mechanism is uniquely suited for middle income countries. Further adoptions of digital-based lending also enables the participation of global financial institutions which may help to facilitate the infusion of foreign technologies through diaspora-based investment and FDI. |
|
M&A Deals: A Way to Combat Innovation Slumps in Pharma
This week we touched on how larger firms are less likely to innovate at a scale compared with smaller firms. The reading Xu (2022) also discussed this, making the point that firms with greater leadership distribution across employees are more likely to innovate rather than developing existing ideas like large firms do. In the biopharma context, I believe this has been addressed partially by the increasing utilization of mergers and acquisitions by which smaller companies with brand new therapeutics or classes of therapeutics are acquired by existing large pharmaceutical corporations. Some high market cap corporations have shifted their entire business model to not having any in-house R&D and relying solely on these acquisitions for their new products. To analyze this trend, I decided to look at the number of newly discovered molecules disclosed to the FDA and how many M&A deals are occurring over the last 4 years (1985-2022). Many pharma companies use the same drugs and put them through phase 4 trials which allows them to gain approval to treat other morbidities, but at the end of the day there is a cap on how much mileage can be gleaned from a certain drug and new molecules must be discovered. Running a pearson correlation provides a value of 0.556 with a p value of 5.68x10-6, which indicates there is a moderate to strong correlation between the number of M&A deals occurring over time and the number of new molecules being discovered. Through this, pharma companies are seemingly able to combat the issue of firms growing too large and becoming stagnant. This is likely because this industry so heavily favors new innovation – new medicines have short terms, typically patent protection lasts 14 years, where they can reap the majority of profits before generics hit the market. From a regulation perspective, the FDA may be able to incentivize innovation by limiting the number of patent-extending trials that companies can apply for. Strategic acquisitions are fueling innovation in the biopharma space, and major contributions can continue to be emphasized over marginal advancements with the right oversight. |
|
From “Tapping into Talent, Coupling Education and Innovation Policies for Economic Growth” we know that having a PhD increases the likelihood of becoming an inventor by 20%. Thus, we would presume that the most productive countries with the highest concentration of innovation would have the highest number of PhD grads, specifically in science and technology. I was unable to find good data on the number of PhD grads for MIC to compare, but below is a graph of PhD graduates in 2019 in 9 countries and their respective proportions in the fields of science and engineering.
Unsurprisingly, the US has the highest proportion of science and engineering PhD grads at 77%, around 43,000 people. China and Japan also have a very high proportion of STEM grads at 75%. In comparison, the percentage of PhDs in STEM fields in Germany, UK, and Canada is more than 10% lower than that of the US and China. This percentage is not insignificant, especially when both the US and China have greater total PhDs. An interesting analysis that could be done with more data would be the number of citations per PhD student in each of these countries, which would be a good indicator of innovative productivity, as well as the proportion of PhD students in these fields who end up in industry rather than remaining in academia. In “Measuring the Characteristics and Employment Dynamics of U.S. Inventors”, we know that over 30% of US inventors are foreign born, with a good majority of those coming from India and China in the past decade. Given the increasing levels of competition between the US and China over technology, one would imagine that a good policy recommendation for the US would be to continue encouraging the attraction of top human capital from these foreign countries. However, a couple weeks ago there was a very heated debate amongst Republicans regarding the H1-B Visa Program and how its overuse has led to foreigners taking jobs from Americans. One of the US’ biggest strengths is its ability to attract top talent from foreign countries into building and innovating in the US. In theory, the foreign PhD grads should be in the top percentiles of their programs, since they are paid the same as American students. Therefore, a decrease in the number of foreign PhDs could potentially also lead to a decrease in the quality (IQ) of PhD students. Moreover, in coming years, the number of Chinese and Indian PhDs will only increase. Assuming that they maintain or increase their levels of productivity per PhD grad, the US will find competition to be increasingly difficult. Hence why it would be unwise for US policy makers to not incentivize the attraction of top foreign talent. |
Innovation and market leadership: Evidence from Russia’s first open firm dataset
In Tuesday lecture, Professor Akcigit discussed the leadership paradox, which suggests that as a firm approaches market leadership, it becomes more politically connected and less innovative (measured by patents per worker). I became interested to explore the relationship between innovation and firms’ market rank in Russia, a developing country. I filtered out companies without information on revenue, investment in intangible assets, and labor expenses. Since the dataset lacked the number of employees per firm, I estimated it by dividing labor expenses by the average industry wage [1](industries were classified according to common accounting standards in Russia). However, wages were not adjusted for regional differences due to unavailable data. I also excluded firms with less than one estimated employee. I calculated investment in intangible assets per labor by dividing investment in intangible assets by the estimated number of employees.
To address this issue, I excluded companies with investment in intangible assets per labor exceeding 30,000 rubles to ensure more reliable results. I then grouped the firms by their market ranks to plot Figure 1. Figure 1 shows a positive relationship between market rank and investment in intangible assets per labor. As firms move away from the market leader in ranking, their innovative activity adjusted for labor tends to increase. The figure, however, also shows greater deviation from the trend line for firms lower in the ranking, indicating that the trend becomes less consistent as firms rank lower. This suggests that while the overall pattern holds, variability in innovation per labor increases as companies fall further from the top ranks. [1] Federal State Statistics Service of the Russian Federation (Rosstat). Labor market, employment, and salaries. Retrieved from https://rosstat.gov.ru/labor_market_employment_salaries |
|
Understanding the age structure of inventors can provide valuable insights into structural barriers to innovation and effectively harnessing talent across diverse age groups. The age of innovative activity is, in essence, a function of factors such as access to resources, opportunity, and societally endorsed or accepted timing for productivity and creative achievement. This memo examines the United States trends of the age distribution of inventors, as reported in the article The Demographics of Inventors in the Historical United States, with a view to the implications for innovation policy and equity.
The figure above presents the ratio of the age of patentees to the average age in the general population between 1860 and 1940. The data indicates that inventors have consistently been older than the general population, with the average age ratio always remaining above 1 throughout the period. Notably, the maximum age ratio peaks around 1900, underlining a historical tendency for innovation to be dominated by more experienced individuals. This trend could indicate that barriers to entry, such as access to education or resources, are skewed toward an older segment of the population rather than the younger talented minds.
The next figure presents a comparison of age distribution by sex and race. Panel (a) shows that among patent filers, women tend to be older than men, perhaps reflecting delayed entry into innovation because of systemic gender biases or career interruptions. Panel (b) shows similar gaps between White and Black inventors, with Black patentees being slightly older. This trend suggests possible structural issues such as reduced mentorship, funding, or networks for underrepresented groups, raising the cost or delaying entry into innovation for those groups. The differences in the age distribution of active patent inventors have profound implications for innovation policy. An increase in young participation through targeting grants, mentorship programs, and challenges will likely raise the rate of innovation as it enhances the likelihood of more fresh perspectives and risk-taking behavior. Similarly, the age gaps among women and underrepresented minority inventors are due to a variety of factors. Policies that reduce systemic barriers - funding disparities, biases in the workplace, or lack of representation - could equalize women and minority groups with their counterparts to contribute to innovation earlier in their careers. These visualizations show the remarkably consistent trends of age-related inequities in the U.S. inventor population. These inequities are driven by systemic inequities in resource access and opportunity, which calls for a targeted intervention of democratizing access to innovation that will lead to an inclusive and dynamic ecosystem. Policymakers can better ensure a more equitable and innovative future by investing in younger, female, and minority inventors. These will not only work on diversifying the pool but also quicken the pace at which technology develops across industries. |
Differential Effects of Education and R&D Subsidies Across Capital-Intensive and Low-Capex IndustriesIn “Tapping into Talent: Coupling Education and Innovation Policies for Economic Growth”, Akcigit et al. examine Danish innovation policy, demonstrating the importance of education subsidies for growth by addressing financial constraints that prevent talented potential researchers from pursuing PhDs. For my memo, I thought it would be interesting to explore a nuance that wasn’t fully addressed—the heterogeneity in capital requirements across different research fields. A key observation from Akcigit et al. is that R&D subsidy effectiveness varies with the relative importance of lab equipment in the research production function. This suggests that R&D subsidies would have differential impacts across fields—likely being more effective for software development compared to fields requiring capital-intensive physical laboratories. Below, I show my initial idea of how to modify the model to differentiate between capital-intensive industries like biotech and materials science (using their original parameters) and low capital expenditure industries like software and theoretical research (with adjusted parameters reflecting lower capital intensity). In particular, I compare broad education subsidies that benefit potential researchers regardless of their eventual field with targeted R&D subsidies specifically for low capital expenditure industries. This comparison explores whether Akcigit et al.'s finding about education subsidies being more effective than R&D subsidies holds even when the R&D subsidies are targeted at industries where they should theoretically have the highest impact (low capex sectors), while the education subsidies remain broad and untargeted. This has important implications for policy decisions where considerations beyond pure economic growth, like national security, might necessitate supporting capital-intensive research fields even if they don't maximize immediate economic returns. The equilibrium equations in the attached LaTeX document show how these different policy approaches affect the threshold ability level and growth rates for each industry type, ultimately determining the overall growth rate of the economy. These equations demonstrate the mechanisms through which targeted versus broad subsidies influence both the intensive and extensive margins of innovation across different sectors.
|
|
“Tapping into Talent: Coupling Education and Innovation Policies for Economic Growth” finds that inventions tend to be concentrated in old and large incumbent firms as almost 63% of inventors work in large firms with 1000 employees at least. In addition to that, super star inventors are less likely to become entrepreneurs as the figure below from the paper demonstrates. Given that smaller teams tend to come with more important innovations that are likely to be more disruptive, as the “Large teams develop and small teams disrupt science and technology” paper points out, it is crucial to understand what is driving this fall in entrepreneurship amongst innovators.
Source: “Tapping into Talent: Coupling Education and Innovation Policies for Economic Growth,” 2024. Ufuk Akcigit, Jeremy Pearce, and Marta Prato. Review of Economic Studies, forthcoming. To assess what is driving the fall in entrepreneurship I suggest the following model: The model considers four variables that are most likely driving the fall in entrepreneurship amongst innovators. The first is the wages inventors earn, as inventors might be incentivized to maintain job stability at larger firms where they earn higher wages. The second is firm size, which as the paper points is largely attracting innovators at increasing rates. The third variable is the reward innovators get from their patents, as high financial rewards received from innovations in larger firms might largely discourage entrepreneurship. |
|
In this memo, I discuss the role that income plays in the decision to get a PhD. Following Akcigit et al (2020), getting a PhD is costly but deepens human capital and prepares for research that gets recirculated in the economy through patents, papers, or other innovations. People are born with endowments of intelligence and other abilities that make them well-suited for research. However, people have preferences and other constraints which control their decision to further invest into their human capital. (Indeed, more recent research has investigated how risk aversion may affect the decision to invest in human capital, as well as possible insurance policies that could smoothen this.) 
A social planner concerned with economic growth faces the issue illustrated in the above Figure 16. There are many people who could be a “good fit” for research, either due to taste or ability. However, to do so, they must both have the taste and willingness to pay (in credit, loans, deferred consumption, etc.) to get a PhD. This mirrors much of the discussion of the decision to pursue postsecondary education, where the role that credit constraints play has been heavily disputed (see Belley and Lochner (2007), Carneiro and Heckman (2002)). Akcigit et al (2020) confirm the result using Danish data that of the people who could research, the majority do not go into research because of taste, not because of financial constraints. I use data from the American Community Survey from 2015 to 2022 and roughly compare schooling decisions with earnings. Without much filtering, PhD holders earn much more than other college graduates (with or without graduate degrees), with about $110,000 average earnings compared to $77,000. Bare-bones statistics are reported below table. 
However, because of the presence of “superstar” earners, it is difficult to visualize their respective income distributions. I attempt to reduce the influence of superstar outliers and other potential black swans by filtering out people making over a million dollars, then narrow my sample to people between the age of 21 and 37. 
In the below histogram, I plot the distribution of income of non-PhD holders, then the income distribution of PhD holders below that. Note that PhD holders’ distributions have a taller “peak” in the $40,000-$60,000 annual salary bin, while non-PhD holders seem to have their center closer to $60,000-$80.000. However, these differences are granular, and PhD owners have a larger median salary than non-PhD holders (75k versus 56k). PhD holders will earn more than non-PhD holders—but at the cost of foregone earnings in the starts of their careers and potentially later into their lives. Adding lifetime earnings would resolve this unintuitive result. However, it suggests credit constraints may not be the full story behind PhD acquisition, which creates a huge problem—what if people just don’t want PhDs, or prefer to be working sooner? 
|
and others
Post your memo in response any (or all) of the week's readings and an empirical case regarding artificial intelligence, innovation, and/or growth:
Post by Thursday @ midnight. By 1pm Friday, each student will up-vote (“thumbs up”) what they think are the five most interesting memos for that session. The memo should be 300–500 words (text) + 1 custom analytical element (e.g., equation, graphical figure, image, etc.) that supports or complements your argument. These memos should: 1) test out ideas and analyses you expect to become part of your final projects; and 2) involve a custom (non-hallucinated) theoretical and/or empirical demonstration that will result in the relevant analytical element. Because these memos relate to an empirical case students hope to further develop into a substantial final project and because they involve original analytical work, they will be very difficult to produce with generative AI and we strongly discourage you from attempting it. Some of the top-voted memos will form the backbone of discussion in our full class discussion and break-out room sessions.
The text was updated successfully, but these errors were encountered: