In England, students apply to universities using teacher-predicted grades instead of their final end-of-school A-level examination results. Predicted rather than achieved grades therefore determine how ambitiously students apply to and receive offers from the most selective courses. The Universities and Colleges Admissions Service (UCAS) encourages teachers to make optimistic predictions to motivate students to apply ambitiously and achieve higher grades. However, little is known about variations in optimism across students and schools, as well as the mechanisms behind such variations. If certain groups of students or schools are predicted more optimistically than others, this may distort application, offer, and acceptance rates between these groups. Such distortions have the potential to impact efforts to promote wider participation and enhance social mobility. In this study, we use newly linked administrative education data to show predicted grades are differentially optimistic by student sociodemographic and school characteristics. These variations are often substantial and can only be partially explained by differences in students' prior achievements, the subjects they studied at A-level, the degree subjects they pursue, and their choices of university and courses. We find less educationally advantaged students are in general more rather than less optimistically predicted, although there are two important exceptions to this trend. Once we control for GCSE score and A-level subject, greater optimism is observed in independent schools and among Oxbridge applications. Thus, differential optimism is positively impacting some of the most educationally advantaged students in the country. Our findings contribute to the growing consensus advocating for reforms to the admissions system, including whereby students can continue to revise their course choices until they receive their achieved grades, and universities only make offers after that date.