Research

This paper studies how changes to firm productivity or size propagate over a production network. It derives an endogenous measure of market power when firms internalise their position in the production network to extract monopoly rents. The paper first explores the relation between network position and endogenous markups under general non-parametric production functions. This separates economically meaningful channels of interaction from restrictions introduced by the choice of functional form of production and the competition regime. It then studies how changes to productivity or firm-size propagate over the network when markups are endogenous. The forward equation of productivity gains and the backward equation of demand creation are complemented with a horizontal equation that captures changes in endogenous market power.
Using granular administrative data from Rwanda and an instrumental variable strategy based on exogenous border closures, the paper tests the empirical validity of the derived measures. It then uses the universe of firm-to-firm interactions in the country to study the network-position of Foreign Direct Investment (FDI) in Rwanda and assess the general equilibrium effects of FDI on the domestic economy via efficiency gains, demand creation, and changes to the structure of competition.

This paper provides evidence of spillover effects from Foreign Direct Investment (FDI) through forward linkages. It analyses granular information on the universe of firm-to-firm transactions and inward FDI in Rwanda. Event-studies reveal substantial and persistent effects on value-added, employment, and productivity of domestic firms after beginning to source from foreign-owned enterprises. These effects are more pervasive than those associated with selling to foreign-owned firms – the backward linkages emphasised in the literature. FDI in Rwanda is motivated by the access to regional markets. This contrasts with the access to cheap input factors for export-driven investment in contexts where spillovers through backward linkages are salient. In a market-seeking context of investment in an input-constrained economy, forward linkages are the dominant channel for spillover effects. Foreign-owned firms provide higher-quality intermediate inputs than domestic suppliers. This applies more strongly to specialised business and professional services that are difficult to import. Inputs sourced from foreign-owned firms complement rather than crowd out domestically sourced inputs.

This paper proposes a synthetic control estimator to identify treatment effects in panel data when units interact over an observed network, treatment is endogenous, and outcomes are subject to network interference. The proposed estimator balances not only pre-treatment outcome trajectories for treated and synthetic control groups, but also their position on the observed network. It does so by adding a balancing objective over a node-similarity kernel based on observed pre-treatment linkages. The network kernel balancing objective improves the synthetic control estimator along three dimensions. First, in settings with few pre-treatment observations the network kernel adds information on unit similarity and improves small sample performance of the estimator. Second, by explicitly balancing the network position of treated and control units the method becomes robust to unobserved shocks that affect outcomes non-linearly as long as these shocks operate through features of the observed network that are well represented by the node-similarity kernel. Third, the network balance implies that treated and control units are exposed to similar levels of indirect treatment via network interference as long as treatment over the network is dense and the kernel function is well-specified. The choice of node-similarity kernel determines the classes of interference captured by the estimator. The paper proposes three examples of kernels and compares their performance to standard synthetic control methods using Monte Carlo simulations.