Cyclical Patterns in the Variance of Economic Activity

This paper models the conditional mean and variance of real GNP and its components using asymmetric exponential generali zed autoregressive conditional heteroskedasticity, a model previously applied only to financial variables. The results im ply that the variance of real GNP is higher following negative innovatio ns than positive innovations and that this asymmetry arises in the cyclically sensitive sectors. Further evidence links this asymmetry to the phase of the business cycle: the conditional variance appears to be largest around business cycle troughs. The evidence of asymmetry in conditional variance is robust to alternative models of output growt h, including split-trend and threshold specifications.