This paper assesses the determinants of market impact costs of institutional equity trades, using unique data from the world's second largest pension fund. We allow the impact of trade characteristics and market conditions on trading costs to depend on the level of trading costs itself and establish significant differences in the responses of cheaper and more expensive trades. We explain the distinct responses from differences in information content and demand for liquidity between trades with high and low trading costs. Finally, to illustrate the practical relevance of the approach, we use our method to forecast future trading costs.market impact costs; quantile regression; forecasting market impact costs