Which young firms grow large and old? We show that the answer depends on how quickly firms learn what they are good at---their product-specific expertise. In our general-equilibrium model of firm dynamics, firms learn faster as they age and with better management and more data. Consistent with the theory, we find in U.S. Census data on manufacturers that young firms select products on estimated expertise about half as sharply as mature firms, and better-managed, more data-intensive firms select more sharply. In the calibrated model, firms that become large and old are mostly those born with high productivity, and among them, those that learn quickly what they are good at. Faster learning amplifies up-or-out dynamics: it raises early exit among firms with weak products but makes survivors more likely to become large. A policy that helps startups learn faster lowers survival but raises consumption by 0.5 percent net of cost, through better production decisions and selection inside firms.

* Previously circulated under the title "Product Switching and Young Firm Dynamics."

(Previous Version: "Defensive Innovation and Firm Growth in the U.S.: Impact of International Trade" [Paper] [Technical Appendix])

In this paper, I show that increasing foreign competition contributed to the recent decline in U.S. business dynamism by changing firms' innovation decisions. Using industry-level regressions with a generalized difference-in-difference identification strategy, I first show that rising competition from China substantially reduced young firm activities, startup rates, and employment growth of high-growth firms in the U.S. in the 2000s. I then develop a two-country model of firm innovation and show that increasing competitive pressure by foreign firms reduces high-growth firm activities and startup rates in the U.S. by inducing innovation-intensive and thus fast-growing firms to focus more on innovation for their product improvement (internal innovation) for defensive reasons. As these incumbents build technological barriers to better protect their markets, firms find it harder to enter others’ markets through innovation for business takeover (external innovation). As a result, the startup rate declines, and firms reduce investment in external innovation for entering new product markets. Since business takeover makes firms grow faster than they would through their own product improvement by requiring firms to hire a new set of workers to produce new products, this change in innovation patterns cuts the employment growth of innovation-intensive firms.

(Previously circulated under the title "Mega Firms and Recent Trends in the U.S. Innovation: Empirical Evidence from the U.S. Patent Data")

- NBER Working Paper No. 31460

We provide evidence that mega firms have played an increasingly important role in shaping new technological trajectories in recent years. While the share of novel patents—patents introducing new combinations of technological components—produced by mega firms declined until around 2000, it has rebounded sharply since then and the share of novel patents in all patent applications in the U.S. also rebounded accordingly. Additionally, we find that the technological impact and knowledge diffusion from novel patents by mega firms have grown relative to those by non-mega firms after 2001. We also examine potential drivers of this trend, presenting evidence that the rise in novel patenting by mega firms is linked to their disproportionate increase in cash holdings and expansion of their technological scope.

(Previous Version: "Competition, Firm Innovation, and Growth under Imperfect Technology Spillovers"  [SSRN])

- IZA Discussion Paper No. 17581

- CES Working Paper No. 24-20 (with previous title)

We study how frictions in learning others’ technology, termed "imperfect technology spillovers," impact firm innovation strategies and the aggregate economy through changes in innovation composition. We develop an endogenous growth model that generates strategic innovation decisions, where multi-product firms improve their products via own-innovation and enter new product markets through creative destruction under learning frictions. In our model, firms with technological advantages intensify own-innovation as learning frictions enable them to protect their markets from competitors, thereby reducing creative destruction of rivals. This pattern gets more pronounced when competitive pressure increases exogenously. Importantly, the shift in innovation composition reduces aggregate growth, as creative destruction contributes more to growth. Using U.S. administrative firm-level data, we provide regression results supporting the model predictions.

In this memo, we construct a patent assignee-firm longitudinal bridge between U.S. patent assignees and firms using firm-level administrative data from the U.S. Census Bureau with web search results. We match granted patents applied between 1976 and 2016 to the U.S. firms recorded in the Longitudinal Business Database (LBD) in the Census Bureau. Building on existing algorithms in the literature, we first use the assignee name, address (state and city), and year information to link the two datasets. We then introduce a novel search-aided algorithm. Overall, we are able to match 88.2% and 80.1% of all U.S. patents and assignees respectively. In particular, the search-aided algorithm significantly improves the matching results by accounting for 7% and 2.9% of matches at the patent and assignee level, respectively. We also find that the match rates accounted for by the search-aided algorithm have substantially increased over time, which highlights its importance in more recent periods. We contribute to the existing literature by 1) improving the match rates and quality with the web search-aided algorithm, and 2) providing the longest and longitudinally consistent crosswalk between patent assignees and LBD firms. (All results reported in the abstract were cleared for release under CBDRB-FY22-P2095-R9872).

This paper constructs a patent assignee-firm longitudinal bridge between U.S. patent assignees and firms using firm-level administrative data from the U.S. Census Bureau. We match granted patents applied between 1976 and 2016 to the U.S. firms recorded in the Longitudinal Business Database (LBD) in the Census Bureau. Building on existing algorithms in the literature, we first use the assignee name, address (state and city), and year information to link the two datasets. We then introduce a novel search-aided algorithm that significantly improves the matching results by 7% and 2.9% at the patent and the assignee level, respectively. Overall, we are able to match 88.2% and 80.1% of all U.S. patents and assignees respectively. We contribute to the existing literature by 1) improving the match rates and quality with the web search-aided algorithm, and 2) providing the longest and longitudinally consistent crosswalk between patent assignees and LBD firms.