Week 8: Memos - Catalyzing Innovation

[jamesallenevans]

Post your memo in response any (or all) of the week's readings and an empirical case regarding artificial intelligence, innovation, and/or growth:

• “Lack of Selection and Limits to Delegation: Firm Dynamics in Developing Countries,” 2021. Ufuk Akcigit, Harun Alp and Michael Peters. American Economic Review, 111(1): 231–275.
• “The Life Cycle Of Plants In India And Mexico,” 2014. Chang-Tai Hsieh and Peter J. Klenow, Quarterly Journal of Economics, 129(3): 1035-1084.
• “Metaknowledge.” 2011. Evans, James A., and Jacob G. Foster. Science 331(6018): 721-725.
• “Science of science.” 2018. Santo Fortunato, Carl T. Bergstrom, Katy Börner, James A. Evans, Dirk Helbing, Staša Milojević, Alexander M. Petersen et al. 2018. Science 359(6379).
• “Toward a more scientific science.” 2018. Pierre Azoulay, Joshua Graff-Zivin, Brian Uzzi, Dashun Wang, Heidi Williams, James A. Evans, Ginger Zhe Jin et al. Science 361(6408): 1194-1197.
• “The diversity–innovation paradox in science.” 2020. Bas Hofstra, Vivek V. Kulkarni, Sebastian Munoz-Najar Galvez, Bryan He, Dan Jurafsky, and Daniel A. McFarland. Proceedings of the National Academy of Sciences 117(17): 9284-9291.

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.

[e-uwatse-12]

Measuring Factors towards Career Progression and Diversity Innovation Paradox

Underrepresented scholars produce disproportionately novel research but receive less recognition and fewer academic opportunities. Investigating this issue is crucial to understanding systemic barriers in academia and designing policies that promote equitable scientific progress. Prior research suggests that underrepresented scholars contribute highly novel ideas but experience systematic discounting in citations, journal acceptances, and career advancement. Investigating these patterns empirically can provide quantifiable evidence of biases and help shape institutional interventions.

CPi = β1INi + β2Ri + β3Di + β4(INi × Di) + β5(Ri × Di) + ϵi

Where:
CP_i = Career progression measure for researcher i (e.g., tenure attainment, faculty position)
INi = Innovation Novelty of researcher i (e.g., unconventionality of citations)
Ri = Recognition received (e.g., citations, journal impact factor)
Di = Demographic indicator (1 if underrepresented, 0 if majority)
INi × Di = Interaction term capturing how innovation affects career progression differently for underrepresented groups
Ri × Di = Interaction term capturing whether recognition disparities exist across demographic groups
β1, β2, β3, β4, β5 = Coefficients to be estimated
ϵi = Error term

Research novelty is often a key driver of scientific progress, yet it may not translate into career benefits equally across demographic groups. Analyzing the relationship between innovation, citation impact, and hiring outcomes can clarify the extent to which undervaluation affects long-term academic success. If biases in recognition and reward systems are found, institutions can implement strategies such as double-blind peer review, mentorship programs, or funding mechanisms to counteract these effects. Findings could inform grant agencies and universities on how to better support high-impact, underrecognized research.

To rigorously investigate this paradox, a proper analysis would focus on Innovation Novelty, using citation networks and text analysis to measure unconventionality in research contributions; Recognition Metrics, examining disparities in citation counts, journal impact, and career progression indicators; Demographic Disparities, comparing how underrepresented versus majority scholars are evaluated given similar levels of novelty; and Network Effects, assessing whether co-authorship with majority-group scholars influences the reception of novel ideas. The expected impact of this study will provide concrete, data-driven insights into how academia values and rewards innovation across different demographic groups. By identifying potential disparities, we can inform policies that foster a more inclusive and meritocratic academic environment. The next steps include finalizing dataset sources and methodological framework, identifying potential collaborators with expertise in bibliometrics and social network analysis, and developing preliminary findings for discussion with institutional stakeholders. Investigating the Diversity-Innovation Paradox is not just an academic exercise—it is a step toward ensuring that all contributions to science and scholarship are equitably recognized and rewarded. Below is a graphical representation illustrating the disparity in citation impact versus research novelty among demographic groups, highlighting the systemic undervaluation faced by underrepresented scholars.

[ypan02]

Analysis of Women Inventors in U.S. Patent Teams

Understanding the representation of women in research and innovation is crucial for fostering diversity and inclusion in intellectual property development. To investigate the presence of women in patent teams, I analyzed a dataset obtained from the U.S. Patent and Trademark Office (USPTO) consisting of 7,747 patents with a publication date of February 18, 2025. By cleaning and transforming the XML dataset into a CSV format, I was able to apply python packages to study gender distribution within patent teams and its relationship with team size and patent classifications.

The dataset contained a column labeled "inventor_names," which I leveraged to determine the gender composition of each patent team using the gender_guesser.detector API. This allowed me to compute the number of women inventors on each patent team and derive the percentage of women within each team. I created scatter plots using matplotlib to explore trends in team composition.

One of the most striking patterns was the relationship between team size and the proportion of women inventors. After removing an outlier team consisting of more than 120 members, I observed a downward trend in the proportion of women as team size increased. The shape of this trend resembled the first half of a quadratic function (x^2), suggesting that larger teams tend to be more male-dominated, while smaller teams exhibit greater variance in gender composition, including some all-women teams.

Further analysis revealed significant gender disparity in single-gender teams. Out of 7,747 patent teams, only 130 were composed entirely of women, whereas 6,385 consisted solely of men. This stark difference highlights a gender imbalance in patent authorship, with all-male teams being nearly 50 times more common than all-female teams.

Running a statistical regression of the percentage of women inventors on team size, I found that the coefficient was statistically significant (p-value = 0.001). This indicates a strong negative correlation between team size and women’s representation, reinforcing the observed downward trend.

To investigate whether gender distribution influences the classification of patents, I performed one-hot encoding on the patent classification column. The dataset comprised 150 unique patent classes. By plotting bar graphs for the top 10 most common patent classes, I discovered that in 7 of these classes, the teams were composed entirely of women. Conversely, when examining the bottom 10 (least common) patent classes, only 2 of them had all-women teams.

This finding suggests that women inventors tend to be more active in patent classes that are widely pursued, rather than in niche or highly specialized fields. The reasons behind this trend could range from industry dynamics to cultural and educational factors that influence women’s participation in specific research areas.

[dishamohta124]

Diversity–Innovation Paradox in U.S. Academia

Research by Hofstra et al. (2020) highlights the "diversity–innovation paradox" in the United States, where scholars from underrepresented groups (women and racial minorities) exhibit higher rates of scientific innovation compared to majority groups yet receive less recognition, citation, and career advancement. This paradox presents a significant challenge for policymakers and academic institutions, which traditionally assume innovation directly leads to career success.

Empirical data spanning three decades (1977–2015) from U.S. doctoral dissertations demonstrates this paradox clearly. Using natural language processing, Hofstra et al. measured innovation as the introduction of novel conceptual linkages (i.e., new associations between previously unlinked scientific concepts). Their results revealed that minority researchers consistently introduced innovations at a higher rate. However, despite this increased innovation, their work experienced lower rates of subsequent adoption by peers, reducing their professional impact.

A core empirical finding is encapsulated by the following analytical equation representing the "impact gap" between majority and minority innovators:

$$\text{Impact Gap} = \frac{1}{N}\sum_{i=1}^{N} (U_{M,i} - U_{m,i})$$

Where:

• $U_{M,i}$ is the average uptake (citations/adoption rate) of novel concepts introduced by majority group researchers in discipline $i$.
• $U_{m,i}$ is the average uptake of novel concepts introduced by minority group researchers in discipline $i$.
• $N$ is the total number of academic disciplines analyzed.

Hofstra et al.'s data showed a positive $\text{Impact Gap}$, indicating systemic undervaluation of minority-introduced innovations across disciplines. For instance, concepts introduced by women had significantly lower uptake compared to their male counterparts, and similarly, racial minority groups faced reduced adoption rates compared to majority peers.

Addressing this paradox requires structural reforms targeting peer review, hiring, and tenure processes. Institutions might consider blinded peer reviews, mandatory citation audits, and diversity quotas on editorial boards. Evaluating the effectiveness of these interventions would involve monitoring changes in the $\text{Impact Gap}$ over time, aiming to achieve parity ($\text{Impact Gap}\approx 0$). Continued empirical evaluation is essential for ensuring that scientific innovation is equitably recognized and rewarded, maximizing academia's overall creativity and societal impact.

[aveeshagandhi]

Firm Dynamics and Growth Constraints in Developing Countries
Akcigit et al. (2021) and Hsieh & Klenow (2014) give us a deep dive into firm dynamics in developing economies. Professor Ufuk and his team argue that weak selection processes and limited managerial delegation prevent full-potential productivity growth, which curtails high-productivity firms from expanding. Hsieh & Klenow, on the other hand, study that plants in India and Mexico exhibit slower growth versus United States, mostly because of the distortions in K and L.

However, this research does intersect at the insight that distortions in firm dynamics —managerial constraints or resource misallocation—lead to lower aggregate productivity. The first paper lays emphasis on weak selection as a barrier to scaling, while Hsieh & Klenow highlight inefficiencies in market frictions. Both of these perspectives suggest that policy interventions should focus on improving managerial capabilities and reducing frictions in labor and capital markets.

Analytical Illustration: Firm Size Distributions
I aim to analyse the above by comparing firm size distributions across countries. The figure below, constructed using empirical estimates, compares firm size densities in India and USA.

Interpretation of the Figure

Key takeaways:

• There are more small firms and fewer large firms: India’s right-skewed distribution
• The U.S. has a more balanced distribution, while still being more skewed to the left, indicating more medium and large firms.
• As firm size increases, density in India sharply reduces, but in the US, density stabilizes for larger firms.
• These trends align with the literature. The dominance of small firms in India shows that there exist some barriers to entry and growth, which backs the argument: firm expansion is limited by selection constraints and market distortions.

Policy Implications

Policymakers should prioritize the following to boost aggregate productivity:

• Better incentives for managerial delegation as well as overall training
• High-productivity firms should have more access to K_t
• Reforming labor and capital markets to allow firms to scale effectively

References
“Lack of Selection and Limits to Delegation: Firm Dynamics in Developing Countries,” 2021. Ufuk Akcigit, Harun Alp and Michael Peters. American Economic Review, 111(1): 231–275.
“The Life Cycle Of Plants In India And Mexico,” 2014. Chang-Tai Hsieh and Peter J. Klenow, Quarterly Journal of Economics, 129(3): 1035-1084.

[kbarbarossa]

Diversity in finance is a subject of intense debate. As a woman pursuing a career in investment banking, I’ve personally experienced the complexities surrounding DEI initiatives. During my internship this past summer, I frequently heard colleagues assume that anyone from an underrepresented background was a "diversity hire," including myself. There was an unspoken belief that DEI programs, rather than merit, were the primary reason for diverse candidates’ presence in the firm. This sentiment wasn’t directed at me alone—I noticed similar assumptions being made about others who benefitted from diversity initiatives, regardless of their qualifications. It made me question whether DEI efforts were truly fostering inclusion or inadvertently reinforcing skepticism about underrepresented individuals in high-stakes fields like finance.

To explore these perceptions on a larger scale, I conducted a sentiment analysis of discussions in finance-related subreddits. The findings are interpreted through Hofstra et al.’s (2020) Diversity-Innovation Paradox in Science, which suggests that while diversity enhances innovation, its contributions are often undervalued in professional and academic settings. My goal was to investigate how online discourse reflects attitudes toward diversity, specifically within finance, particularly given the recent backlash against DEI in corporate settings.

Using Python’s PRAW (Python Reddit API Wrapper), I scraped posts and comments from finance-related subreddits that discussed diversity. To ensure a balanced perspective, I searched for keywords reflecting both support and criticism, including "DEI," "affirmative action," "meritocracy," "reverse discrimination," and "equity." Sentiment analysis was conducted using the VADER lexicon, which classified text as positive, neutral, or negative.
An analysis of over 700 posts and comments revealed a stark divide in attitudes toward diversity in finance:
-70% of discussions were positive, emphasizing DEI’s role in broadening perspectives and fostering inclusion.
-15% were negative, raising concerns about meritocracy, tokenism, and claims that diversity initiatives undermine hiring standards.
-15% were neutral, consisting of factual discussions or ambiguous opinions.

Hofstra et al. (2020) argue that diversity fuels innovation, yet diverse contributors often face skepticism or devaluation. This paradox is reflected in online discourse—while many recognize the benefits of diverse teams in decision-making and risk assessment, there remains resistance to DEI initiatives. Much of the criticism aligns with Hofstra et al.’s findings: even when underrepresented voices contribute valuable insights, their legitimacy is questioned.

A key insight from this analysis is the contrast between corporate decisions and online sentiment. While many companies are scaling back DEI policies, discussions in finance-related communities remain overwhelmingly supportive of diversity initiatives. This raises an important question: If public sentiment toward DEI remains largely positive, what is truly driving these corporate reversals?

Ultimately, this analysis highlights the complexity of fostering diversity in professional environments. Striking a balance between benefitting from diverse perspectives and maintaining a meritocratic system is not straightforward. As attitudes toward DEI continue to evolve, understanding the nuances behind these discussions is crucial for shaping policies that promote both fairness and effectiveness.

[spicyrainbow]

How does DEI programs fail to address the diversity-innovation paradox
In the article “The Diversity–Innovation Paradox in Science”, we learned that in academia, while diverse minority groups introduce more novel conceptual linkages and are more likely to innovate, however, their innovation are often undervalued and less frequently adopted. which led to lower offer of career rewards than similar contributions from white men. This devaluation stems from structural biases that make it harder for women and racial minorities to receive recognition and advance in academia.

Recently, there has been growing opposition to DEI programs across Fortune 500 companies, universities, and even in the White House. Many organizations argue that these initiatives fail to drive meaningful business growth and have instead become bureaucratic burdens. However, as the article demonstrates, diversity does enhance innovation. The question, then, is why have DEI programs failed to drive the business growth its suppose to do?

One of the key struggles underrepresented innovators face is having their work valued. Yet, research has shown that mandatory DEI programs often reinforce biases and discrimination instead of reducing them. When individuals feel forced to accept certain ideas, they tend to push back to assert their autonomy. This resistance deepens bias rather than eliminating it, meaning DEI programs fail to address the core challenge that minorities face—getting their innovations recognized and rewarded, which stops their ideas to be implemented or chance to further innovate in higher position.

Moreover, the Diversity–Innovation Paradox is not just about individual biases—it is about institutional structures that fail to reward the contributions of marginalized scholars. Underrepresented researchers must produce significantly higher levels of novelty just to receive the same career recognition as their majority-group counterparts. Yet, many DEI programs were originally designed not to foster true inclusion but to protect companies from discrimination lawsuits.

If the underlying goal of DEI programs is not to promote real diversity and drive innovation, then leadership will not invest in actively elevating underrepresented team members or championing their ideas. True innovation from diverse groups cannot be fully realized if there are no real efforts to integrate these individuals into leadership roles. Instead, many minority employees remain concentrated in lower levels of the team hierarchy, where their innovative perspectives are overlooked, and they have limited opportunities to contribute to ideation and decision-making, the cycle worsens.

Ultimately, DEI programs fail because they focus on optics rather than substantive structural change. Without shifting the goal from avoiding lawsuits to internalizing the true benefit of innovation from diverse teams, innovation from diverse groups will decrease, be undervalued, and the systemic barriers that perpetuate the Diversity–Innovation Paradox will worsen leading to cancellation of DEI programs that we saw recently.

I created a theorized visualization to show how ineffective DEI program leads to the cycle of DEI failure and broke down the goals of a successful DEI program needs to achieve in order to help address the diversity-innovation paradox to truly benefit and drive business growth from diversity innovation production.

[sijia23333]

MEMO: A Dynamic Framework for Delegation-Driven Growth

I propose a new Dynamic Delegation-Growth Framework that captures how managerial delegation affects firm growth potential:

$$ G_{f,t} = \theta_f \cdot \alpha^{\sigma n_f} \cdot e^{-\phi(n_f - n^*)} \cdot \left(\frac{M_f}{T_f}\right)^\beta $$

Where:

• $G_{f,t}$ represents firm f's growth potential at time t
• $\theta_f$ is the firm's inherent type
• $\alpha$ is the delegation efficiency parameter
• $n_f$ is current firm size
• $n^*$ is the delegation threshold size
• $M_f$ is outside managerial input
• $T_f$ is owner's time endowment
• $\sigma$, $\phi$, $\beta$ are elasticity parameters

Key Novel Mechanisms:

1. Dynamic Threshold Effects: The exponential term $e^{-\phi(n_f - n^*)}$ introduces a new way to model how growth constraints evolve as firms cross critical size thresholds. This captures non-linear scaling effects missing from standard growth models.

2. Multiplicative Type-Delegation Interaction: By multiplying $\theta_f$ with delegation efficiency raised to firm size ($\alpha^{\sigma n_f}$), the model reveals how inherent firm capabilities and delegation efficiency interact dynamically rather than additively.

3. Resource Intensity Scaling: The managerial intensity ratio $(\frac{M_f}{T_f})^\beta$ provides a novel measure of how effectively firms leverage outside talent relative to founder constraints.

Empirical Applications:

The equation generates several testable predictions:

1. Growth discontinuities should appear around $n^*$, with distinct scaling behaviors above and below this threshold. Using detailed firm-level panel data, we can identify these transition points through regression discontinuity designs.

2. The model predicts that firms with similar $\theta_f$ values but different $\frac{M_f}{T_f}$ ratios will display divergent growth paths after crossing $n^*$. This can be tested by comparing matched pairs of firms that differ in their managerial hiring patterns.

3. The multiplicative structure implies that improvements in delegation efficiency ($\alpha$) should have exponentially larger effects on growth rates as firm size increases, creating testable predictions about the distribution of firm growth rates conditional on size.

Conclusions:

The framework could be particularly valuable for understanding why some firms successfully scale while others stagnate, even when they appear similar on observable characteristics. It suggests that seemingly small differences in delegation efficiency can compound into large growth divergences through the interaction effects captured in the equation.

[mskim127]

The Korea Discount: Structural Barriers to Shareholder Value and Innovation in South Korean Conglomerates

South Korea’s equity markets have suffered from a persistent valuation gap compared to global peers, a phenomenon widely termed the “Korea Discount.” This discount reflects systemic corporate governance failures within the country’s dominant family-controlled conglomerates, or chaebols, whose strategies prioritize retaining familial control over maximizing shareholder value. This misalignment of incentives has created a self-reinforcing cycle of capital misallocation, suppressed innovation, and diminished competitiveness. By hoarding cash, avoiding shareholder-friendly policies, and engaging in opaque related-party transactions, chaebol leadership stifles mechanisms that drive long-term growth.

The Korea Discount stems primarily from the outsized influence of chaebols, which account for approximately 60% of the nation’s GDP. These conglomerates, including Samsung, Hyundai, and LG, are structured to ensure familial control through cross-shareholdings and pyramid ownership schemes. Controlling shareholders often prioritize empire-building and intergenerational wealth preservation over minority investor returns, leading to chronic undervaluation. For instance, chaebols frequently retain earnings rather than distribute dividends, with South Korea’s dividend payout ratio remaining among the lowest in emerging markets. This practice not only depresses stock prices—reducing inheritance tax liabilities for founding families—but also limits capital available for productive reinvestment.

The lack of minority shareholder protections exacerbates this dynamic. South Korea lacks mandatory takeover rules common in developed markets, enabling chaebols to acquire controlling stakes in subsidiaries at inflated prices before squeezing out minority investors at discounted valuations. Such tactics erode investor confidence, necessitating higher risk premiums and perpetuating the valuation gap. Empirical studies indicate that chaebol-affiliated firms trade at a 15–30% discount relative to non-chaebol peers, even after adjusting for sector and profitability.

Chaebols’ tendency to hoard cash–often characterized as a buffer against economic cyclicality–has dire consequences for innovation. Retained earnings, which reached approx $900 billion in 2023 across listed firms, are frequently deployed into low-return ventures or unrelated subsidiaries to consolidate control rather than fund R&D.

The inheritance tax regime further distorts incentives. With rates as high as 60%, founding families actively suppress valuations to minimize tax burdens during generational transitions. For example, LG Group’s 2022 restructuring involved transferring stakes in LG Chem to family holdings at below-market prices, a move that diluted minority shares but preserved familial control. Such maneuvers divert resources away from innovation-intensive sectors, entrenching reliance on legacy industries like shipbuilding and petrochemicals.

[anishganeshram]

Capital Intensity of R&D in the Biopharmaceutical Sector

Recent data highlights substantial growth in industry-sponsored clinical trial starts alongside rising expenditures on research, suggesting that biopharma remains one of the most capital-intensive fields in global innovation. As illustrated in the first figure, the number of industry-sponsored interventional clinical trial starts has climbed from around 3,000 in 2012 to almost 5,000 in 2021 and 2022, while inflation-adjusted pharmaceutical R&D spending has likewise trended upward. Although trial activity has expanded, probabilities of success from Phase 1 to 3 hover between 10 percent and 19 percent, underscoring the risk inherent in drug development.

The second figure offers additional perspective on innovation outputs by tracking the number of granted patents from 2017 to 2021. Overall patent grants (the yellow line) initially rose from 5,813 in 2017 to 6,868 in 2020 before receding to 6,204 in 2021. Within this total, grants at the USPTO increased from 4,257 in 2017 to a peak of 5,018 in 2020 but then fell to 4,688 in 2021. Meanwhile, EPO grants climbed from 1,556 to 1,850 between 2017 and 2019, but they also declined in the last two years of available data. These patterns suggest that the global patent landscape for pharmaceuticals can fluctuate in response to regulatory shifts and investment cycles.

Capital intensity in biopharma arises partly from the need to finance high-cost clinical trials, sophisticated lab infrastructure, and specialized human capital. The substantial fixed costs of scientific research—together with the need for large-scale collaborations among scientists, clinicians, and manufacturing partners—drive up the total expense of bringing a therapy to market. As the broader literature on scientific production notes, additional resources are often required to manage big data, comply with regulatory standards, and protect intellectual property through patent filings. In short, scientific knowledge generation in this industry has significant financial requirements, reflecting a high degree of risk and complexity.
Furthermore, the high failure rate of drug candidates—even after considerable capital investments—illustrates the importance of understanding implicit biases, research heuristics, and institutional structures that shape decisions about which targets to pursue. This resonates with the concept of “metaknowledge,” which emphasizes how the production of knowledge depends on both explicit data and the contextual factors behind scientific work. Funding levels, institutional prestige, and collaboration networks all influence a project’s direction and chances of success.

Biopharmaceutical R&D epitomizes the capital intensity of modern science. Rising trial starts and shifting patent grants indicate sustained activity and competition, but they also demonstrate the substantial resources required to develop new therapies. The inherent uncertainty surrounding clinical outcomes demands large upfront investments, which can only be recouped when a successful drug attains regulatory approval and market acceptance. Ongoing analysis of these trends should help stakeholders allocate resources more effectively and refine policy incentives, ultimately improving the efficiency of the innovation process in life sciences.

[cskoshi]

Measuring the effect of AI on the optimal manager $m(n)$

In the paper “Lack of selection and limits to delegation”, we are given the objective function that managers have to solve:

This gave us the equation for the “delegation cutoff $n *(\alpha)$, which is

I wanted to play around with this equation by bringing in the effects that communication technology, such as that of zoom or google drive, would have on the cut off at which a firm decides to hire managers, and cross reference this with data on change in the number of managers across the US.

My initial hypothesis was that with the introduction of communication technologies like discord, zoom, etc, entrepreneurs human capital T would increase, as they are better able to “be at multiple places at once” in order to delegate work. Before the technological advances of recent times, T was constrained by things such as geographical distance, which proved a hindrance to their ability to get work done. This severely limited their human capital as even if they had the energy to do so, it was prohibitively difficult to literally manage across timezones and continents. With the advances and mass adoption of these technologies, it now becomes easier to multitask and manage. Additionally, it would also raise the efficiency of outside managers, which is given by $\alpha$. The same reasons apply here, managers who are hired are better able to manage due to the removal of barriers such as geographical differences. I hence took this as an exogenous shock to T and $\alpha$. This presents an interesting problem, as they have counteracting effects on the m(n) term. To better understand this, I wanted to take a look at the partial derivatives of both T and $\alpha$. I got these:

$$ \frac{\partial m(n)}{\partial \alpha} = \left( \frac{\sigma}{\omega_M} \right)^{\frac{1}{1-\sigma}} \cdot \frac{\sigma}{1-\sigma} \cdot \alpha^{\frac{2\sigma - 1}{1-\sigma}} + \frac{T}{n \alpha^2}. $$

$$ \frac{\partial m(n)}{\partial T} = -\frac{1}{\alpha n}. $$

For simplicity’s sake, let’s assume that the elasticity of profit with regards to T is >0.5 (ie $\sigma > 0.5$, because if not, the effect is indeterminate which makes it more fun but also requires more than a memo to understand). First, we see that $\alpha$ is positively related to m, so if $\alpha$ increases, so does m. Conversely, if T increases, m decreases. This makes intuitive sense as if managers become more productive, the optimal level to hire would increase, and if the entrepreneur has less limited capital, then I’d hire less managers optimally. Given that (\alpha) is less than 1, (\frac{1}{\alpha n}) can be quite large, especially if (n) is small. Therefore, the magnitude of (\frac{\partial m(n)}{\partial T}) is likely to be larger than that of (\frac{\partial m(n)}{\partial \alpha}) in many cases, especially when (\alpha) is small and (n) is not too large. In short, we’d expect that the optimal manager amount decreases due to this exogenous shock.

Indeed, while we do not see a decrease, we do in fact see a fall in growth rates of managers over time. I used numbers from Forbes and the Bureau of Labor Statistics to get these. Admittedly, as they are from different sources they might have different classifications but the trend is generally well reported.

In conclusion, it’s interesting to see the implications of this model on the number of managers that firms overall would hire. Whilst we see the biggest harm of technology on employment as that of direct replacement, we should also note the less immediate effects, such as the dampening need for these middle managers. This also brings up the idea of equity, as technology advances, who stands to gain? More and more throughout the quarter, we see that it’s the entrepreneurs and those on top that can use technology to further entrench their position not only within the market but within the firm.

[jacksonvanvooren]

AI Patent Growth in the US vs. India

Hsieh and Klenow in “The Life Cycle of Plants in India and Mexico” find that the surviving plants in the United States grow significantly faster than those of India and Mexico. The latter tends to have firms double employment, while in the US, a similar firm would grow its labor force sevenfold. Akcigit, Alp, and Peters in “Lack of Selection and Limits to Delegation: Firm Dynamics in Developing Countries” attribute this slower growth to the lack of managerial delegation–firms in India often rely on family and not outside managers.

Today, India has a booming technological sector and is a pioneer in AI research. Moreover, these corporations are increasingly multinational. Using patent growth as a proxy for the aggregate growth of AI corporations, this memo explores whether we see similar trends as those in the articles above. The primary goal is to examine patent growth patterns in the US and India.

Methodology

I merged patent, assignee, location and abstract data, then filtered for a set of AI keywords (including “deep learning”, “neural networks”, etc.) in the abstracts. I filter for US granted AI patents and India granted AI patents. Note these are patents filed with the USPTO from companies in India. This will likely bias the results to large, multinational corporations compared to the smaller, family-managed companies as described in the paper by Akcigit, Alp, and Peters.

I calculate the percent change in patents for each year between 2007 and 2023. This normalizes patent growth instead of examining raw counts $(n_{USA} \approx 30,000 \text{ vs. } n_{India} \approx 10,000)$.

Results and Discussion

First, I calculated the proportion of companies in the USA and India that have not filed a patent since 2019. This serves as a potential proxy for a company not surviving. We find that in the United States, 29% of AI related companies filed one or more patents before 2019 but none in 2020 until today. The number for India is much higher, at 50%. This supports Hsieh’s and Klenow’s argument that the life cycle of plants in India tends to be shorter, generally speaking. I also examined year-over-year patent growth and rolling averages to smooth out the graph.

India exhibits higher variance in AI patent growth, both in raw percentage change and the smoothed changes. The United States, on the other hand, has had a steady increase in yearly patent growth and its growth is strictly positive. Still, India has had years with higher average AI patent growth than the US, which demonstrates potentially inconsistent, but larger growth.

India’s AI firms (such as Infosys or Haptik) are not managed by families and instead have deep, hierarchical organizational structures. Given this, and that AI is a thriving industry in both the US and India, it is not surprising to observe similar patent growth rates. Though AI is relatively new, the similar growth in companies across both countries points to some potential convergence in innovation dynamics.

Code here.

[Adrianne-Li]

Memo: The Role of AI in Scientific Discovery and its Impact on Innovation Trajectories

Introduction

The advent of artificial intelligence (AI) in scientific research is reshaping knowledge production by altering how hypotheses are generated, how literature is synthesized, and how new discoveries emerge. This memo investigates the extent to which AI-driven discovery influences the balance between exploration (novel idea generation) and exploitation (refinement of existing knowledge). Using a simulated dataset, we employ econometric models to analyze citation patterns and knowledge diffusion dynamics, providing empirical insights into the implications of AI in scientific progress.

---

Empirical Framework

The relationship between AI-driven research influence and knowledge diffusion is modeled using a panel dataset of scientific papers. The following empirical model is estimated:

$$Citations_{i,t} = \beta_0 + \beta_1 AI_{i,t} + \beta_2 Novelty_{i,t} + \beta_3 AI_{i,t} \times Novelty_{i,t} + \beta_4 Collaboration_{i,t} + \delta_t + \gamma_i + \epsilon_{i,t}$$

Where:

• Citations_{i,t} represents the number of citations received by paper i at time t.
• AI_{i,t} is a binary indicator for whether the paper was AI-assisted.
• Novelty_{i,t} captures how conceptually distant the paper is from prior work (measured via text embedding similarity scores).
• AI_{i,t} \times Novelty_{i,t} interacts AI usage with novelty to measure differential AI effects on novel research.
• Collaboration_{i,t} controls for co-authorship network effects.
• δ_t and γ_i are time and paper-specific fixed effects, respectively.
• ε_{i,t} is the error term.

---

Regression Results

                            OLS Regression Results                            
==============================================================================
Dep. Variable:              Citations   R-squared:                       0.442
Model:                            OLS   Adj. R-squared:                  0.438
Method:                 Least Squares   F-statistic:                     98.13
Date:                Tue, 25 Feb 2025   Prob (F-statistic):           1.93e-61
No. Observations:                 500   AIC:                             1408.
Df Residuals:                     495   BIC:                             1429.
Df Model:                           4                                         
Covariance Type:            nonrobust                                         
==========================================================================================
                             coef    std err          t      P>|t|      [0.025      0.975]
------------------------------------------------------------------------------------------
const                      2.1373      0.154     13.869      0.000       1.835       2.440
AI_Assisted                1.5843      0.174      9.101      0.000       1.242       1.926
Novelty                    1.9222      0.206      9.333      0.000       1.517       2.327
AI_Novelty_Interaction    -1.2825      0.302     -4.247      0.000      -1.876      -0.689
Collaboration              0.5187      0.040     12.925      0.000       0.440       0.598
==============================================================================

---

Key Findings from Data Analysis

1. AI-Driven Research Diffuses Faster but is Less Novel

   • Papers using AI-assisted research methodologies exhibit higher short-term citation rates.
   • However, AI-generated discoveries are less likely to introduce conceptually distant ideas compared to traditional research approaches.

2. Interaction Between AI and Novelty

   • AI enhances the diffusion of ideas, but the positive citation effect diminishes when research is highly novel.
   • This suggests a potential crowding-out effect, where AI reinforces mainstream knowledge rather than pushing the frontier of scientific discovery.

3. Collaboration Networks Influence AI Adoption

   • Research groups with strong interdisciplinary ties are more likely to integrate AI in their methodologies, yet their citation impact remains subject to existing field norms.

---

Visualizations

Figure 1: AI Influence on Citation Rates

Figure 2: AI’s Effect on Novelty-Adjusted Citations

Figure 3: Interaction Effect of AI and Novelty on Citations

---

Policy Implications and Conclusion

The findings suggest that while AI accelerates knowledge diffusion, it may also reinforce existing paradigms rather than fostering breakthrough innovations. Policymakers and funding agencies should consider:

• Incentivizing AI applications that explicitly target high-novelty research.
• Supporting interdisciplinary collaboration to counteract potential knowledge concentration effects.
• Refining evaluation metrics to ensure that AI-driven research does not disproportionately favor incremental over radical innovation.

Future research should further investigate the causal mechanisms behind AI’s influence on innovation trajectories, leveraging natural experiments and exogenous shocks in AI policy adoption.

References

• Evans, J.A., & Foster, J.G. (2011). Metaknowledge. Science, 331(6018), 721-725.
• Fortunato, S., et al. (2018). Science of science. Science, 359(6379).
• Hofstra, B., et al. (2020). The diversity–innovation paradox in science. PNAS, 117(17), 9284-9291.

---

[michelleschukin]

Foreign vs. Domestic Ownership and Firm Innovation in Emerging Markets

The paper Lack of Selection and Limits to Delegation: Firm Dynamics in Developing Countries by Akcigit, Alp, and Peters examines why firms in developing economies tend to remain small and less productive compared to those in developed economies. A key argument is that managerial constraints limit firm growth, preventing the emergence of highly productive, innovative firms. This paper highlights the role of selection effects, where inefficient firms survive due to a lack of competitive pressure, and delegation limits, where firm owners struggle to scale operations due to weak managerial structures.
To explore this idea, I analyzed firm ownership patterns using the World Bank Enterprise Survey dataset. Specifically, I examined the distribution of domestically-owned and foreign-owned firms in a mix of developed and developing countries to assess whether foreign investment correlates with larger, potentially more innovative firms.
Analytical Element: Firm Ownership and Its Implications for Growth
The bar chart below illustrates the proportion of domestically-owned and foreign-owned firms across ten countries, including Brazil, China, France, India, Korea (Rep.), Lebanon, Mexico, Poland, Russia, and Türkiye. These countries were selected based on their relevance to the study’s themes: some represent rapidly growing economies with strong foreign direct investment (FDI) (China, Mexico, Poland), while others (India, Lebanon) have economies characterized by a high prevalence of small, domestically-owned firms.

Findings:

1. India has an overwhelmingly high number of domestically-owned firms, reinforcing the argument that developing economies struggle with firm expansion. This supports the paper’s idea that weak delegation and managerial constraints prevent firms from scaling.
2. Developed economies (France, Korea, Poland) have a more balanced distribution of foreign and domestic firms, which suggests that foreign investment plays a role in fostering competitive, larger firms.
3. Foreign-owned firms are a small proportion in all countries, but their presence may signify better managerial structures, which are associated with firm expansion and higher productivity.

Conclusion: Implications for Innovation and Policy
The findings from the dataset reinforce the paper’s argument that managerial constraints and weak competitive pressure limit firm expansion in developing economies. The dominance of domestically-owned firms, particularly in countries like India and Lebanon, suggests that scaling challenges and barriers to foreign investment may stifle innovation and productivity growth.
In contrast, a more balanced ownership distribution may indicate that foreign investment brings better managerial structures, access to global markets, and potentially greater innovation spillovers. While foreign-owned firms remain a minority across all countries, their presence could signal higher firm efficiency and scalability, as suggested by the paper's delegation and selection constraints framework.
A key policy question emerges: Should developing economies actively encourage foreign investment to overcome managerial limitations and accelerate firm growth? Future research could explore whether foreign-owned firms introduce more innovation and whether their management practices can be replicated by domestic firms to enhance competitiveness.

[yhaozhen]

Recent work on the “Diversity–Innovation Paradox in Science” suggests that demographic diversity spurs higher novelty in research, but underrepresented scientists do not receive proportional credit or career rewards for that novelty. The central finding is that women and racial minorities produce more “uncommon” links between ideas (proxied by “new conceptual linkages”), yet these contributions gain less uptake. One underappreciated angle involves the way AI might amplify this paradox. Current AI-driven fields (e.g., NLP, machine learning) often feature gatekeepers who define “standard tasks,” “benchmark datasets,” and “performance metrics.” Underrepresented scholars can propose fresh angles—like merging linguistic analyses with clinical data or integrating non-English corpora—yet their ideas might remain on the fringes. If top labs prefer well-trodden tasks or data, these more-distal proposals can be overlooked.

Such barriers could worsen the paradox: on paper, AI is more open, since open-source code and pre-trained models are ubiquitous. In practice, major labs command huge compute resources and have built brand power with specific tasks and leaderboards. When underrepresented researchers introduce original tasks—like merging AI fairness and minority-language text—those efforts might see low citation or replication. A large language model might then “absorb” these new ideas but assign minimal credit or visibility back to the originator, who is outside the influential AI networks.

Addressing this paradox requires more than encouraging “diversity hires.” Funding agencies and conference committees could better highlight offbeat or “distal” AI papers, ensuring that high-risk proposals by newcomers don’t vanish. Or, top AI research labs might add on-ramps for researchers from smaller labs who want to experiment with creative data, ensuring real cross-pollination. The paper’s finding that “impactful novelty” yields less return for underrepresented scientists implies AI’s inclusive potential can go unrealized if citation and hiring practices remain entrenched. Collecting field-specific data on AI publications—tracking citations, open-sourcing patterns, or acceptance rates—will be crucial. Ultimately, if AI is to thrive, it must treat novel viewpoints from diverse researchers not as peripheral or niche, but rather as essential expansions of the discipline’s conceptual landscape.