The complexity of problems related to the formation of cognitive models of complex systems based on methods of econometric analysis of available time series of domain parameters is considered. An approach is proposed to solve the problem of representing a complex system in the form of an understandable and practical model that includes a limited number of factors, but as adequately as possible reflects the dynamic processes occurring in this system. An example of the application of the proposed approach is considered, exploring the possibilities of improving the efficiency of the functioning of financial markets. The vector autoregression model, which is a model of the dynamics of several time series, is chosen as the basis for the formation of the model. Options for improving the model by adding virtual factors and cointegrating relations are proposed, and the results of the study of the developed model are also presented