Probabilistic forecasting guides dynamic decisions

Real-world assets---such as projects, jobs, skills, or relationships---change endogenously with continued investment, often improving at variable rates. These structural dynamics pose a challenge: How do we decide what asset to invest in and when to divest from an underperforming asset? In this paper, we propose a normative model of asset selection and switching. We hypothesize that people use Bayesian function learning to infer invariant, structural parameters underlying changes in asset performance, and choose assets based on forecasted final performances at a task-specific time horizon. We test the model in four experiments ($N = 460$) that manipulate the setup (selection versus switching), time horizons, and statistical properties of asset performance trajectories. The Bayesian forecasting model consistently outperforms myopic alternatives inspired by foraging and study time allocation, which instead assume direct continuation of recent observations and rely solely on them. Our results highlight a distinctive form of future-looking cognition: humans anticipate how their own actions will shape future asset performance, and draw on representations of underlying structural dynamics to inform their current choices. The Bayesian probabilistic forecasting model therefore offers a principled computational account of how humans make decisions with lasting consequences.