We formulate a price discovery model in which the price discovery measures vary either locally, say, for instance, at intervals of 30 minutes or at a daily frequency. Given the empirical and theoretical evidence that price discovery measures relate to highly persistent fundamentals, we adopt a kernel-based approach that allows parameters to vary smoothly over time. The resulting kernel-based price discovery measures are consistent and asymptotically normal. Empirically, we compute daily estimates of price discovery for 30 stocks in the U.S. over a long time span, from 2007 to 2013. We show that there is significant daily variation in the component share measures with relative market informativeness alternating between the listing and competing exchanges.
