Should Macroeconomists Consider Restricted Perceptions Equilibria? Evidence from the Experimental Laboratory

This paper studies a simple model of output and inflation in the experimental laboratory. While the Rational Expectations Equilibrium (REE) predicts output and inflation to be white noise processes, output and inflation in experimental sessions display stable cyclical patterns. For about 50 model periods agents' expectations, which are the sole source of these patterns, are described extremely well by a Restricted Perceptions Equilibrium (RPE). In this equilibrium agents use the univariate forecast function which generates the lowest mean squared forecast error at the 1-step forecast horizon and iterate these forecasts to derive multi-step predictions. After about 50 model periods agents seem to learn that their simple univariate forecast function is misspecified and start to employ different forecast models for different prediction horizons. The data suggests that the different models are again optimal univariate forecast functions and evidence in favor of convergence towards the REE remains weak, even after more than 100 model periods. However, for model parameterizations where an RPE does not exist, agents' expectations are captured relatively well by the REE