Expectations of heterogeneous economic agents play a pivotal role in modern macroeconomic theory. Since the standard assumption about a representative fully rational agent and his/her ability to form model-consistent expectations of the underlying process governing real economic outcomes is subject to well-grounded criticism, laboratory experiments are an important tool for gaining new knowledge about the formation of individual expectations. This paper provides an overview of the results of learning-to-forecast experiments based on the New Keynesian model that allow to identify specificities of agents’ expectations formation. These results suggest specific recommendations on ways to increase efficiency of the Central Bank monetary policy. Special attention is paid to agents’ inflation expectations and specificities of applying various types of inflation targeting. The article presents an analysis of the impact of announcing the inflation target on agents’ inflation expectations and the main macroeconomic variables dynamics for two different inflation targeting regimes — strict and flexible. A comparison of point and band versions of inflation targeting is provided for shocks affecting the economy with different intensities. The results of learning-to-forecast experiments presented in the review provide evidence of training and switching between different forecasting heuristics and also indicate the presence of specific cognitive restrictions of agents that extend a certain heuristic rule to all macroeconomic variables for which the forecast is performed.