## Cognitive Consistency and Signal Extraction in Models of Adaptive Learning

A guiding principle for the adaptive learning approach of bounded rationality is “cognitive consistency,” which requires that learning agents be no more knowledgeable than real-life economists. Much of current research falls short of satisfying this principle because agents are often assumed to observe the exogenous vector of the economy. The key assumption of this paper is that agents observe the same information as econometricians: they cannot observe fundamental shocks and can only observe a finite sample of past endogenous variables. They must extract information from available data using statistical techniques. In a new Keynesian model, multiple equilibria arise even when the monetary policy satisfies the Taylor principle. Learning can be used to eliminate some equilibria, but there can still be more than one learnable equilibrium. Model dynamics are also significantly different when agents use simple statistical models to forecast the future.

Download file here.

## Leave a Reply