Issue about relevancy of usage of concepts of representative and aggregate agents inmo-dern economic science is very actual. In theoretical model [Malakhov, Pospelov, 2014]showed, that distribution of banks on shares of assets is stable over time. If this results is correct for real data, then it will another argument to usage of concept of aggregate agent in modeling of banking sector, which is an actual topic for macroeconomists. In this paper we provide an empirical test of this result using data from Russian banking system. We also analyze other key variables, such as households’ deposits, firm’s credits, interbank credits, etc., because if distributions of shares of these variables are stable too, then it will be additional argument to usage of concept of aggregate agents. Aim of this paper is selection of optimal (in some sense) functional forms of distribution of shares of key variable and validating stability of these distributions over time. Actuality of this topic is also confirmed by recent events in Russian economy and banking system in particular.

We show that using generalized versions of well-known distributions, we can accurately describe the distribution of Russian banks in terms of turnover balance sheet. In particular, the Pareto distribution of type IV and asymmetric generalized error distribution show a very high accuracy of approximation and these results are correct for all considered variables. Quality of approximation by these distribution is robust, both in time and in the cross-sectional dimension, however, individual banks can move in distribution. Thus, we can’t talk about the distribution of individual banks but of the distribution of banks of the entire Russian banking system.

Moreover, estimations of parameters of distribution of shares of assets have been minorly changing during observation period and these changes could be possibly connected with structural shifts in banking industry. Kolmogorov – Smirnov test shows, that differences between distributions of shares of assets become significant at 5% confidence level, only when differ ence between periods is more than 8 months. Thus theoretical model [Malakhov, Pospelov, 2014] mainly passes the empirical test.