While optimistic and pessimistic predictions are distinct dimensions, their combined patterns and longitudinal associations with mental health remain underexplored. This study examined two indices of future thinking: optimistic prediction (OP), defined as the difference in likelihoods between positive and negative events, and general expectation (GE), the average of these likelihoods. Because various combinations of OP and GE are theoretically possible, we adopted cluster analysis to determine the actual prediction patterns found in our data. An online survey was conducted involving 888 participants at Time 1, of whom 449 completed an 8-month follow-up at Time 2. The participants provided likelihood predictions for 12 positive and 12 negative events across three temporal horizons: 1 month, 1 year, and 10 years. Cluster analysis identified four patterns: positive-dominant, high-certainty, low-certainty, and negative-dominant. Dependent variables, including hopelessness, depression, life satisfaction, and trait anxiety, were examined across these patterns. Chi-square tests showed that these patterns remained significantly stable over 8 months. In addition, moderated regression analyses and structural equation modeling demonstrated that GE accounts for unique variance in hopelessness independently of OP. Furthermore, GE × OP interaction revealed that low GE strengthens the association between low OP and hopelessness. Cross-lagged models revealed a stable relationship between future expectations and hopelessness over time, suggesting that these constructs are interrelated. Specifically, the low-certainty pattern (low GE) was associated with greater hopelessness, comparable to the negative-dominant pattern. These findings suggest that clinical interventions may benefit from being tailored to individuals' prediction profiles.