|
WORKING PAPER
Sensitivity of GARCH Estimates: Effects of Model Specification on Estimates of Macropartisan Volatility
Gleditsch, Kristian S.Maestas, Cherie Abstract This paper explores the volatility of aggregate partisanship using a generalized autoregressive conditional heteroskedasticity (GARCH) model of the variance. We are particularly interested in how different specifications of the mean model affect the variance estimates. Modeling the variance of macropartisanship is theoretically interesting because such a model can capture periods of greater and lesser volatility in aggregate party identification. However, given the widespread debate over the dynamic properties of the aggregate partisanship time series, a range of plausible specifications for the mean model should be considered before drawing conclusions about variance estimates. We find similar estimates of the variance effects using ARMA-GARCH, ARFIMA-GARCH, ARIMA-GARCH and ECM-GARCH models. Weak ties to party consistently predict greater volatility in all four models, while presidential election quarters are associated with greater volatility in three of the four models. Counter to our expectations, the candidate centered era of the last few decades is associated with lower average variance. Finally, all four models indicate that volatility tends to persist beyond the duration of the shock that sparks it. Keywords ARCH/GARCH volatility of aggregate partisanship File
Uploaded 05-24-1998 Document ID Number 291 |
|
|