Working Paper: CEPR ID: DP15330
Authors: Fabio Canova
Abstract: I examine the properties of cross sectional estimates of multipliers, elasticities, or pass-throughs when the data is generated by a conventional multi-unit time series specification. A number of important biases plague estimates; the most relevant one occurs when the cross section is not dynamic homogenous. I suggest methods that can deal with this problem and show the magnitude of the biases cross sectional estimators display in an experimental setting. I contrast average time series and average cross sectional estimates of local fiscal multipliers for US states.
Keywords: cross sectional methods; dynamic heterogeneity; partial pooling; fiscal multipliers; monetary passthrough
JEL Codes: E0; H6; H7
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| dynamic homogeneity (C69) | valid estimates of dynamic effects (C51) |
| dynamic heterogeneity (C69) | biases in estimating average dynamic effects (C22) |
| cross-sectional methods (C21) | significant biases in estimating average dynamic effects (C22) |
| government expenditure (H59) | output (C67) |
| cross-sectional methods (C21) | zero or negative fiscal multiplier (E62) |
| alternative methods (Q42) | estimates statistically indistinguishable from one (C13) |
| choice of estimation method (C51) | change in policy conclusions (F68) |