Unemployment is notoriously difficult to predict. In previous studies, once country and year fixed effects are added to panel estimates, few variables predict changes in unemployment rates. Using panel data for 29 European countries collected by the European Commission over 444 months between January 1985 and October 2022 in an unbalanced country*month panel of just over 10000 observations, we predict changes in the unemployment rate 12 months ahead. We do so using individuals' fears of unemployment which predict subsequent changes in unemployment 12 months later in the presence of country fixed effects and lagged unemployment. We also use industrial firm's expectations of future employment, which are also predictive of what happens to unemployment three months later. Using our preferred model specification, we present out-of-sample predictions based on replications from 1,000 random samples. These track actual movements in unemployment rates closely over a period in which there were two major recessions and unemployment shifted by a factor of two.