Adaptive agents may be smarter than you think : unbiasedness in adaptive expectations

Agents forming adaptive expectations generally make systematic mistakes. This characterization has fostered the rejection of adaptive expectations in macroeconomics. Experimental evidence, however, shows that in complex environments human subjects frequently rely on adaptive heuristics – model-consistent expectations being simply too difficult or impossible to implement – but their forecasting performance is not as inadequate as assumed in the characterization above. In this paper we show that adaptive agents may not be as gullible as we used to think. In a model with adaptive expectations augmented with a Belief Correction term (which takes into account the drift of the macroeconomic variable of interest) the average forecasting error is frequently close to zero, hence (belief amended) adaptive expectations are close to unbiasedness.