Learning in Overlapping Generations Models

We consider a standard two generations version of the overlapping generations model with different learning algorithms. Agents predict inflation rates on the basis of some mis-specified (linear) perceived law of motion, which is estimated by running a regression on past prices or inflation rates. Beleifs might converge, although inflation rates keep on fluctuating. These fluctuations are, in some sense, consistent with the (mis-specified) limit belief.