Working Paper: NBER ID: w29908
Authors: Kilian Huber
Abstract: Large-scale shocks directly affect some firms and households and indirectly affect others through general equilibrium spillovers. In this paper, I describe how researchers can directly estimate spillovers using quasi-experimental or experimental variation. I then argue that spillover estimates suffer from distinct sources of mechanical bias that standard empirical tools cannot resolve. These biases are particularly relevant in finance and macroeconomics, where multiple spillover channels and nonlinear effects are common. I offer guidance on how to detect and overcome mechanical biases. An application and several examples highlight that the suggested methods are broadly relevant and can inform policy and multiplier calculations.
Keywords: No keywords provided
JEL Codes: C13; C2; D5; E0; E51; G0; G21; G30; L2; R11; R23; R51
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| large-scale shocks (E32) | some firms and households (G59) |
| large-scale shocks (E32) | others through general equilibrium spillovers (F69) |
| omitting one type of spillover (C24) | bias estimates of another (C51) |
| mismeasurement of treatment status (C32) | mechanical bias in spillover estimates (C21) |
| direct estimation of spillovers (C13) | calculate impact of policy and multiplier effects (C54) |
| traditional model-based approach (C52) | hard-to-verify assumptions about general equilibrium channels (D52) |