Does the level of government matter for industrial policy? I compare firms invested by local government, central government, and private venture capital (VC) funds in China’s semiconductor industry. Central and private VCs invest in the nation's most productive firms, while local VCs back smaller, less productive local champions, suggesting misallocation for the national government maximizing industry sales. Yet event studies suggest that local government-backed firms experience larger post-investment gains of about 27% in revenue productivity relative to not yet invested firms, while private- and central government- backed firms see no detectable change. As local government VCs hold a smaller portfolio and are sit closer to their portfolio firms, they participate more in management by adding directors with graduate degrees and foreign education. Board changes account for 42% of the central-local productivity effect gap. In a model combining VC selection, firm productivity and capital responses, and import competition, the sales of domestic semiconductor firms in 2023 would have been 12% lower had private VCs allocate government VCs' deals, and 4% lower if the central government allocates local governments' investments. Dynamic productivity growth from local VC investments quantitatively outweighs the selection of central and private VCs on initially larger and more efficient firms.
As semiconductors dominate recent industrial-policy debates, understanding their underlying economics is critical. In contrast with the existing literature that relies on indirect estimates of semiconductor manufacturing costs, this paper combines comprehensive product specification data with a commercial-grade engineering cost model, constructing a granular panel of over 77,000 CPU and GPU chip-quarter observations from 2015 to 2025. By reverse-engineering the engineering cost model through a series of regressions, we document several findings that challenge standard modeling assumptions: (i) vintage-capital effects are highly heterogeneous and can even increase costs; (ii) physical learning-by-doing confers a modest 1.9% quarterly cost reduction; and (iii) static scale economies flatten above a 70% capacity utilization threshold. We also find accrual-based depreciation accounting to be a critical source of biases that could dominate all of these estimates if not properly isolated from physical, cash-flow-based costs. Finally, we confirm the presence of persistent firm-level heterogeneity, which could be a major bottleneck for industrial subsidies.
This paper examines how large industrial policy programs shape firm political spending. We argue that firm-level lobbying rises when large, firm-specific policy rents are at stake, especially when the total size and allocation of benefits remain uncertain. Undistributed benefits also induce firms that had never lobbied to overcome fixed entry costs and begin investing in politics. Event studies of two landmark 2022 industrial policies, the CHIPS and Science Act and the Inflation Reduction Act, show that semiconductor firms increased lobbying expenditures by about 52% between 2018 and 2022, while green-energy firms increased lobbying by about 112%. In both industries, new lobbying firms (the extensive margin) and previously active firms (the intensive margin) each contribute substantially, with entry accounting for about 40% of the semiconductor increase and roughly two-thirds of the green-energy increase. The increase persists after enactment, reflecting firms' concerns over ex post subsidy allocation and policy uncertainty from political transitions.