This paper discusses an application of a rather novel technique for the estimation of the tails of return distributions for financial assets. This extreme value approach proves to be particularly useful when assessing characteristics of high frequency (tick-by-tick) transaction data. Tail parameter estimates allow derivation of probabilities of large price changes. These probabilities improve optimal setting of bid-ask spreads based on the order processing component of the bid-ask spread. Estimates for optimal levels are compared to 'observed' bid-ask spreads. The latter, which are estimates in itself, are based on recently developed methods in the literature.