We model the conditional distribution of high-frequency financial returns by means of a two-component quantile regression model. Using three years of 30 minute returns, we show that the conditional distribution depends on past returns and on the time of the day. Two practical applications...