I propose a state-space approach to decomposing a stock's idiosyncratic volatility (IV) into a common component and an idiosyncratic one. The IV of a stock is defined as the stochastic volatility of its idiosyncratic return. Unlike a simple average of IV's in the cross section or the first principal component of IV's in a panel in existing studies, the common component here is an AR(1) process and captures the persistence of IV's at the daily frequency. Using a panel of stocks in the Dow Jones Industrial Average Index from 1963 to 2009, I estimate the model with a Bayesian Gibbs sampling procedure. The model is found to fit the panel of IV's in sample better than a GARCH(1,1) model and a principal component approach. It also forecasts future levels of IVs better than a GARCH(1,1) model in the medium- to long-run. Instead of taking a static view of IV, I assess the pricing implications of the unpredictable part of the common component in the cross section of stock returns at the daily frequency. I find that the common component is not priced in my sample and sub-samples.