This paper studies stock market volatility in Korea using a semiparametric
single index volatility model, in which a single index long run component
induced by exogenous covariates is multiplied to a GARCH short run
component. When a covariate is nonstationary, i.e. integrated or nearintegrated,
the model can account for time-varying unconditional variance of
financial time series. Among various economic and nancial indicators, it is
found that the coincident composite index, VKOSPI, and housing price index
are helpful in fitting and forecasting stock market volatility in Korea. It is
shown that the model using these three variables outperforms standard models
both in terms of in-sample fitting and out-of-sample forecasting.