By Evan Beard
This paper examines the effect of macroeconomic variable volatility on implied and realized asset price level volatilities in the U.S. using monthly data from 1986 – 2008. Two approaches are taken: An autoregressive distributed lag model using rolling standard deviations and a GARCH model. The S&P 500’s volatility is used as a proxy for historical (actual) volatility and the VIX is used as a proxy for implied volatility. For the distributed lag model, each linear regression tests granger causality (using Newey-West robust standard errors) of a single macroeconomic variable by incorporating lagged values (as determined by comparing Bayesian Information Criteria of both the constructed macroeconomic variable and the dependent asset volatility variable). Capacity utilization, PPI, and employment volatility are found to be significant for predicting S&P volatility, while PPI and M2 volatility are significant for the VIX. For the GARCH regressions, terms of trade, employment, and capacity utilization volatility are statistically significant. Forecasts are then constructed using those variables shown to be granger casual, but a two-sided t-test rejects the null hypothesis that forecast errors are zero in every case.
Advisor: Lori Leachman
