In sticky price models based on micro evidence that each period a fraction of prices are kept unchanged, recent studies reach the qualitatively equivalent conclusion that higher trend inflation is a more serious source of indeterminacy of rational expectations equilibrium, regardless of whether labor is firm-specific or homogeneous. This paper shows that the model with firm-specific labor is more susceptible to indeterminacy induced by high trend inflation than the model with homogeneous labor, because these two different specifications of labor lead to distinct representations of inflation dynamics. In addition, the model with firm-specific labor is more susceptible to expectational instability of the equilibrium caused by high trend inflation.

If private sector agents update their beliefs with a learning algorithm other than recursive least squares, expectational stability or learnability of rational expectations equilibria (REE) is not guaranteed. Monetary policy under commitment, with a determinate and E-stable REE, may not imply robust learning stability of such equilibria if the RLS speed of convergence is slow. In this paper, we propose a refinement of E-stability conditions that allows us to select equilibria more robust to specification of the learning algorithm within the RLS/SG/GSG class. E-stable equilibria characterized by faster speed of convergence under RLS learning are learnable with SG or generalized SG algorithms as well.

This paper examines a class of interest rate rules that respond to public expectations and to lagged variables and considers varying levels of commitment that correspond to varying degrees of response to lagged output. Under commitment the policymaker adjusts the nominal rate with lagged output to impact public expectations. Within this class of rules, I provide a condition for nonexplosive and determinate solutions. Expectational stability obtains for any nonnegative response to lagged output. Simulation results show that modified commitment is best under least-squares learning. However, in the presence of parameter uncertainty and/or measurement error in the policymaker's data on public expectations, the best policy is one of partial commitment, where the response to lagged output is less than under modified commitment. The case for partial commitment is strengthened if the gain parameter in the learning mechanism is high, which can be interpreted as the use of few lags by public agents in the formation of expectations or as an indication of low credibility of the policymaker. The appointment of a conservative central banker ameliorates these concerns about modified commitment.

This paper explores the equilibrium properties of boundedly rational heterogeneous agents under adaptive learning. In a modified cobweb model with a Stackelberg framework, there is an asymmetric information diffusion process from leading to following firms. It turns out that the conditions for at least one learnable equilibrium are similar to those under homogeneous expectations. However, the introduction of information diffusion leads to the possibility of multiple equilibria and can expand the parameter space of potential learnable equilibria. In addition, the inability to correctly interpret expectations will cause a “boomerang effect” on the forecasts and forecast efficiency of the leading firms. The leading firms' mean square forecast error can be larger than that of following firms if the proportion of following firms is sufficiently large.

This paper analyzes the relationship between the expectational stability of rational expectations solutions and the possible convergence of adaptive learning processes. Both concepts are used as selection criteria in the case of multiple rational expectations solutions. Results obtained using recursive least squares lead to the conjecture that there exists a general one-to-one correspondence between these two selection criteria. On the basis of a simple linear model and a stochastic gradient algorithm as an alternative learning procedure, it is demonstrated that such a conjecture would be incorrect: There are cases in which stochastic gradient learning converges to rational expectations solutions that are not expectationally stable.