WAR is not actually predictive of future performance, or at least it does not try to be (it is more predictive than WPA though). WAR is measuring value without considering game context - it is a long run stat that assumes outcomes even out by the end of the season.
Predictive stats are regressed and much more conservative, WAR is reporting actual on field performance. Trout's 10+ WAR last year did not mean he should be expected to reach 10 WAR again this year, it simply means his performance last year, on average, should've been worth 10 wins. Interestingly he also lead the league in WPA.