Saddlepath learning

Saddlepath learning occurs when agents learn adaptively using a perceived law of motion that has the same form as the saddlepath relationship in rational expectations equilibrium. Under saddlepath learning, we obtain a completely general relationship between determinacy and e-stability, and generalise minimum state variable results previously derived only under full information. When the system is determinate, we show that a learning process based on the saddlepath is always e-stable. When the system is indeterminate, we find there is a unique MSV solution that is iteratively e-stable. However, in this case there is a sunspot solution that is learnable as well. We conclude by demonstrating that our results hold for any information set.