We study the UK's university application system, in which students apply based on predicted examination grades, rather than actual results. Using three years of UK university applications data we find that only 16% of applicants' predicted grades are accurate, with 75% of applicants having over-predicted grades. However, high-attaining, disadvantaged students are significantly more likely to receive pessimistic grade predictions. We show that under-predicted candidates are more likely to enrol in courses for which they are over qualified. We conclude that the use of predicted rather than actual grades has important implications for labour market outcomes and social mobility.