In this paper we use an autoregressive conditional intensity approach to estimate local high-frequency volatility, and examine to what extent a large universe of market microstructure variables affects local volatility. Our findings support a sequential information arrival hypothesis on the high-frequency level since we show that contemporaneous trading volume is negatively, and lagged trading volume is positively related to local volatility. The use of a penalized likelihood method allows us to obtain a ranking in terms of the relative importance of all market microstructure variables considered. We find that, in a descending order, contemporaneous volume, bid-ask spread, absolute order imbalance, absolute order flow and absolute quote difference carry the most important information for local volatility modelling.