A. R. Bakhtizin, N. V. Bakhtizina

The study presents an approach to construct hierarchical dynamic models with macro- and micro level agents. It aims at modeling the behavior of these agents more realistically than currently existing methods. Our approach is based on building hybrid agent-based models, where neural networks are used to formalize the reaction of households on macro system-s signals. Note that agent-based modeling extends traditional modeling. Agent-based modeling may partially include other approaches of imitational modeling. The latter may be employed "inside" an agent-based model when certain agents or objects of the model are formalized. The emergence of agent-based models may be regarded as evolution of modeling methodology: the change from mono models (one model - one algorithm) to multi models (one model - a set of independent algorithms). The methodology we suggest for constructing hybrid agent-based models includes the following stages: 1) Formulation of the goal for modeling the behavior of households. 2) Implementation of a special sociological survey for obtaining data on the reaction of micro level agents on changes in macro economic situation. If the data of existing surveys monitoring the behavior of Russian households (e.g., RLMS or NOBUS) contain enough variables for the set goal, it may also be employed. 3) Specification of the behavior of micro level agents by constructing corresponding neural networks. 4) Formulation of a set of micro level agents which represent the components of an agent-based model or, in other words, an "artificial society". 5) Realization of an economic system in the form of computable general equilibrium model (macro level medium), which determines "medium for functioning" of artificial societies. On the basis of the developed methodology we created a hybrid model for socio-economic system of Russia. Macro economic agents in the model are grouped according to ownership. Micro economic agents in the model are individuals, who make their decisions at the job market.