On the Properties of Various Estimators for Fiscal Reaction Functions

IMF Working Papers, July 1, 2006

Source details

Canonical URL: On the Properties of Various Estimators for Fiscal Reaction Functions

Evaluates the bias of the least-squares-with-dummy-variables (LSDV) method in fiscal reaction function estimations.
Fiscal reaction functions examined: relationships between the primary fiscal balance and its determinants, including public debt and the output gap.
Identifies a previously unexplored methodological issue: lagged debt is not a strictly exogenous variable, which biases the LSDV estimator in short panels.
Approach: derive the bias analytically to understand its determinants and run Monte Carlo simulations to assess its likely size in empirical work.
Main conclusion: the bias of LSDV in fiscal reaction functions is smaller than the bias of the LSDV estimator in a comparable autoregressive dynamic panel model; LSDV outperforms a number of alternatives in estimating fiscal reaction functions.

Analytical derivation of the bias of the LSDV estimator when lagged debt is endogenous in short panels.
Monte Carlo simulations used to assess the likely magnitude of the bias in empirical applications.
Comparison framework: contrast LSDV bias in fiscal reaction function specifications with LSDV bias in autoregressive dynamic panel models; evaluate LSDV performance against several alternative estimators.

Lagged debt is not a strictly exogenous variable in fiscal reaction function estimations, creating bias in LSDV estimates in short panels.
The analytically derived bias clarifies the determinants of the LSDV bias in this context.
Monte Carlo evidence indicates:
- The bias of LSDV in fiscal reaction functions is smaller than the bias of LSDV in comparable autoregressive dynamic panel models.
- LSDV outperforms a number of alternative estimators for the fiscal reaction function setting.