Research

This paper studies the impact of artificial intelligence (AI) adoption on workplace misconduct. Using regression analysis and quasi-natural experiment around the open-source launch of Google Brain's TensorFlow machine learning toolkit, we find that firms with higher AI intensity experience significant and persistent declines in workplace violations and penalty amounts. The effects operate primarily through productivity-enhancing complementarities and discretionary expenses increases, while labor-adjustment channels play no role. Benefits are concentrated among larger, intangible- and organizational-capital intensive firms, which highlights uneven gains from AI adoption and suggests that AI may widen compliance inequality among firms.

This paper studies how the interaction between intangible capital and skilled labor shapes firm-level productivity. We proxy skilled labor using firm-level artificial intelligence (AI)-skilled workers and exploit the 2015 release of Google TensorFlow as a plausibly exogenous shock to AI effectiveness. Using a difference-in-differences framework, we show that firms with higher pre-shock AI exposure increase their intangible capital, and these increases translate into higher labor productivity. Moreover, productivity gains are concentrated in large firms, while smaller firms show little response. This heterogeneity implies that the interaction between intangible capital and AI workers contributes to rising productivity dispersion across firms.

We study how artificial intelligence (AI) affects firms’ sales in the pre-ChatGPT era. Exploiting the 2015 release of Google TensorFlow as a plausibly exogenous shock to AI-worker effectiveness, we implement a difference-in-differences design that leverages firm-level variation in pre-shock AI exposure. We show that AI adoption generates large and persistent increases in total, domestic, and foreign sales, but only among large firms; smaller firms exhibit no responses. Mechanism evidence suggests that while AI exposure raises operating efficiency and productivity for large firms, firms with financial constraints and limited intangible capital are less able to translate AI into sales gains. 

Artificial intelligence (AI) is reshaping how new ideas are produced, but AI must itself be invented, and inventing it takes a scarce input, the workers who can build AI systems. Does a firm's stock of AI labor cause it to invent more AI? Using firm-level data on AI hiring and AI patenting for U.S. public firms, I exploit the November 2015 open-sourcing of TensorFlow, a large and external drop in the cost of building deep-learning systems, interacted with firms' pre-shock AI-labor exposure. Firms that entered the shock with more AI labor produced more AI patents afterward. The effect loads on deep-learning technologies, and is driven by large and organizational capital rich firms. It operates through rising R&D and a growing number of AI inventors, and it makes AI invention more concentrated among the firms that were already AI-rich. An instrumental-variables strategy that instruments AI labor with a university-graduate supply shifter confirms a positive, AI-specific effect. 

We study the relationship between corporate board characteristics and firm-level artificial intelligence (AI) adoption. Linking BoardEx director data to Compustat for U.S. public firms, we find at the board level that firms whose directors have larger external networks, more qualifications, shorter tenure, and more diverse nationalities adopt substantially more AI than their industry and year peers, and that this relationship is strongest in small, young, high-growth, and AI-intensive firms. The same profile reappears at the CEO level: CEOs with larger networks, more qualifications, and shorter tenure run firms that adopt more AI. We then analyze why some boards adopt AI more. A board-network shift-share design shows that firms adopt more AI when their directors sit on other, more AI-intensive boards, which we read as evidence of a director-borne diffusion channel. Lastly, we use board-member deaths as a quasi-random shock to board composition and instrument the board index with a signed death shock; the component of board AI-orientation moved by these deaths maps into higher AI adoption. 

This paper examines how executive compensation influences firm investment, intangible capital, and innovation by exploiting the elimination of performance-based pay deductibility under the 2017 U.S. Tax Cuts and Jobs Act. Using difference-in-differences and event-study designs, we show that firms more exposed to the shock significantly reduce R\&D, intangible capital, capital expenditures, and patent applications, especially among growth-oriented and smaller firms. We trace these real effects to weaker incentives: high-exposure firms cut stock-based and non-equity incentive pay and experience declines in risk-taking incentives, and they shift toward safer financial policies with higher payouts, lower cash flow, and higher earnings per share.

We construct a novel establishment-level dataset combining the nineteenth-century U.S. Census of Manufactures (1850-1880) with patent records from CUSP to examine how innovation shaped firm performance in early industrial America. Patenting establishments were substantially larger than comparable non-patenters across all dimensions (employment, capital, output, and output per worker) with premia further amplified among firms that produced breakthrough inventions. Estimating establishment-level production functions using Levinsohn-Petrin approach, we find that patenting is associated with higher revenue total factor productivity (TFPR) while leaving markups unchanged on average. Breakthrough innovations, however, are associated with gains on both margins, with a size decomposition revealing the strongest amplification among the largest establishments, consistent with superstar dynamics, yet small breakthrough innovators also capture meaningful gains, suggesting the returns to invention were not exclusive to large firms. 

This study focuses on estimating the role of intangible capital on firms’ competitiveness. We use Lyft’s acquisition of Motivate, the biggest bike sharing company in the U.S. at the time, to evaluate the degree to which intangible capital affects the competition between Lyft and Uber. By acquiring Motivate, Lyft gained more consumer data as we interpret intangible capital, and bikes’ presence on the streets potentially helped Lyft build stronger brand salience. We estimate the effect of the acquisition on Lyft’s ridership by employing trip-level ride sharing data from New York City and using a difference-in-difference-in-differences model. We find that the acquisition helped Lyft increase its ridership by around 6%.

This paper examines how internally generated intangible capital shapes merger patterns and post-merger performance in the U.S. banking sector. We construct a novel measure of intangible capital using granular regulatory expense data and quantify assortative matching between acquirers and targets. Employing a difference-in-differences design with propensity score matching, we causally show that higher assortative matching in intangible capital leads to significant improvements in post-merger bank performance. We complement the empirical analysis with a dynamic search-theoretic model of bank mergers, demonstrating that strategic complementarities in intangibles give rise to assortative matching equilibria. Our findings provide new insights into banking consolidation.

We study the effects of the Reserve Bank of India’s 2006 Bank Authorization Policy on credit allocation, capital productivity dispersion, and firm outcomes using a quasi-natural experiment. We show that private-sector bank branches increased by 16.3% in underbanked districts, while public-sector banks did not exhibit a comparable expansion. Private-sector lending increased disproportionately toward firms with high ex-ante marginal revenue product of capital (MRPK), leading to a decline of approximately 60% in their MRPK and a reduction in district-level MRPK dispersion. These findings emphasize the role of bank ownership in shaping MRPK dispersion by accounting for firm heterogeneity.

We study how non-rival intangible capital interacts with borrowing structure and financial frictions to shape firm dynamics over business cycles. We show: (i) the positive and significant association between intangible-capital growth and labor productivity growth becomes smaller in recessions; (ii) the non-rivalry of intangible capital is evident such that intangible growth predicts faster sales growth and broader firm scope, yet this relationship declines in recessions; (iii) intangible-intensive firms carry less total and secured debt, and substitute toward earnings-based covenant (EBC) borrowing over asset-based covenant (ABC) borrowing; and (iv) intangible-intensive firms with EBC have tightening financially constraints in recessions, which mitigates the productivity payoff of non-rival intangibles. We rationalize these patterns in a general-equilibrium model in which firms draw EBC/ABC constraints at entry and intangibles are non-rival in the firm production technology. The model yields a credit-amplification mechanism with heterogeneous borrowing types, reconciling the productivity slowdown despite rising intangibles.

Corporate intangible capital and nonbank financial institutions (NBFIs) have both risen sharply in the U.S. economy. Intangible-intensive firms, however, are known to face financing frictions where intangible assets make poor collateral and are dif ficult for traditional banks to lend against. We ask whether intangible-intensive firms systematically sort toward nonbank lenders, and, if so, what happens on the loan side in terms of spreads, maturity, and collateral. For U.S. publicly listed borrowers, we show that intangible-intensive firms migrate toward NBFIs, and that this migration is costly such that they pay higher spreads and face stricter collateral requirements when they borrow from nonbanks. A heterogeneity anal ysis reveals a price-quantity asymmetry across firm size, with large firms sorting extensively into nonbank credit while small intangible-intensive firms face more restricted access yet pay the largest rate premiums. We use the U.S. Basel III Final Rule, a regulatory tightening associated with a larger role for nonbank lenders, to examine how the intangible-NBFI relationship is shaped around such a reform. Difference-in-differences estimates show that high-intangible firms shifted dis proportionately toward nonbank credit relative to low-intangible firms, and that, while they took on more nonbank credit, they also faced higher loan rates.