Understanding how individuals form expectations is critical for dynamic economic modeling, particularly in asset pricing. This paper introduces a novel experimental learning-to-forecast design (LtFE), in a positive-feedback expectations environment. In each round, participants provide both one-period-ahead and two-period-ahead forecasts, thereby enabling a more direct examination of how intertemporal expectations evolve. This design extends the traditional two-step-ahead framework by explicitly incorporating the previously implicit intermediate one-step prediction. Preliminary findings reveal the emergence of bubbles, indicating that participants fail to converge to the Rational Expectations Equilibrium (REE). This outcome also facilitates direct comparison with earlier two-step-ahead experiments, shedding further light on how participants form multi-period expectations. Additionally, we propose a double-step forecasting heuristic model that formalizes participants’ decision-making processes under multi-period uncertainty, offering deeper insights into the mechanisms behind expectation formation. Crucially and in contrast to traditional models, our approach captures an agent’s dual mindset capacity to foresee volatile outcomes in the immediate term yet anticipate potential stability in the future, showcasing a duality in expectation formation not accounted for in standard models.