We propose a machine learning methodology for predicting the future liquidity distribution of individual bonds in the U.S. corporate bond market and use it to compute two forward-looking illiquidity measures: expected illiquidity and expected tail illiquidity as measure for downside liquidity risk. We find that bonds characterized by higher expected illiquidity have elevated systematic risk premiums, whereas expected tail illiquidity is predominantly reflected in the alpha. Investors in corporate bond funds preemptively sell their shares in response to anticipated liquidity declines in underperforming funds. All effects are much stronger compared to the standard approach of using today’s realized liquidity.