Spillovers Under Common Shocks in Panel Data: Estimation and Inference
Abstract. Economic outcomes reflect both private effects and spillovers across firms, industries, or regions. Distinguishing these effects is difficult when interaction networks are unknown and unobserved common shocks affect both outcomes and covariates. A high-dimensional spillover network must then be distinguished from co-movement generated by common shocks. This paper develops estimation and inference for bilateral spillover and common private effects with unknown networks and unobserved common shocks. Unobserved common shocks affect outcomes through interactive fixed effects. Identification exploits a factor structure in covariates and exogenous idiosyncratic variation under sparsity conditions. Combining factor projection with debiased LASSO and post-selection OLS, we establish asymptotic normality and simultaneous spillover inference while accounting for factor-estimation and regularization errors. An application to industry-level R&D in the United States illustrates the importance of accounting for common shocks: the number of significant spillover links falls by 42%, and the estimated private R&D elasticity declines and becomes statistically insignificant.