In this paper we integrate big data into the computer model of an arti cial society. The model is agent-based and consists of several modules, representing demographic, economic, nancial processes, employment and consumption, educational and administrative institutions. In order to create an arti cial society that would simulate Russian Federation in 2014 year we use big open data, including Federal State Statistics Service yearbooks and o cial information on the websites of the ministries. Algorithm of an arti cial society reconstruction includes creation of agents and organizations; distributing them among geographical regions; setting interrelations between agents, households and organizations. For the veri cation of the arti cial society model we propose a DES-analysis method, which compares demographic, economic and social indicators of the simulation output data with the real values of these indicators in the base year. We present statistical analysis of modeling results variation in retrospective period.
In the practice of socio-economic systems' management, there is a need to create and implement new methods and tools for making prognosis and planning. Sustainable economic and technological growth requires development of infrastructure, production capacities, human resources and living standards of the population. To take into account multiple factors, it is necessary to analyze big data, including both open statistical information, results of sociological surveys, monitoring of federal programs, and private data from ministries, departments, social networks and search systems. In order to use these data we need special methods and tools. Particularly, we integrate big data analysis methods into computer models of socio-economic processes.
We have chosen agent-based modeling as a main method in this study, since
it allows to re ect dynamics of a macro-system as a result of the interaction of
microlevel objects. The concept of agent-based modeling was proposed in the
1990s [8] and since then has been widely applied in analysis of economic,
nancial, social and environmental processes [
The aim of our research is to construct an agent-based computer model of an
arti cial society, which re ects age and sex structure and regional resettlement
of population, composition of households, economic structures, administrative
and educational institutions. For the information content of the model, federal
statistical yearbooks and o cial information on the websites of the ministries of
the Russian Federation are used [
The developed model includes 7 interconnected modules re ecting various aspects of an arti cial society: Demographics, Education, Employment, Production, Consumption, Financial System and Public Administration (see Fig. 1). Each module corresponds to a set of information objects and events that change their state.
The module \Demographics" re ects birth, maturation and death of agents. New households are formed after marriages and divorces. Agents can act as labor, tax-payers, consumers, creditors and students, thus interacting with the environment and with each other [13]. We set network connections between the agents and determine closeness of the connection in the range from 0 to 1 using the following rules: 0 { not acquainted; 0.1..0.3 { neighbors, colleagues; 0.4..0.6 { friends; 0.7..0.9 { relatives. Relationships among agents are re ected in a square matrix (Table 1). Closeness of the relationship 1 agent has only to himself (the main diagonal of the matrix). Connections are transitive: if the rst agent does not have direct connection (closeness is 0) with the third agent, but has connection with the second agent (e.g. 0.5) and the second agent is connected with the third agent (e.g. 0.8), closeness of the indirect connection between the rst and the third agent will be equal to the product of the intermediate connections (described in example 0.5 * 0.8 = 0.4). Indirect connections are important for disseminating information process. Information about the agent will be available for another agent, if the relationship between them (direct or indirect) is higher than the threshold [14].
Agents A1 A2 : : : Ai : : : An
The model includes three types of organizations: commercial, nancial and budgetary. Organizations interact with individual agents within hiring employees, paying wage or ring them. Beyond that, educational organizations recruit students, promote them to the following courses and graduate them (module \Education"). Financial organizations give credits to agents and commercial organizations and take deposits from them (module \Finance").
Organizations permanently interact with counterparties within sales, deliveries and nancial settlements for them; each operation is re ected in the accounting, which is a simpli ed version of the system adopted in the Russian Federation. Operations are written-o into the table \Accounting entries" as a set of the following structure:
<Date, Deb acc, Deb start, Cred acc, Cred start, Sum, Deb n, Cred n>, where Date { date of the operation, Deb acc { debit account of the entry, Deb start { value of the debit account before the operation, Cred acc { credit account of the entry, Cred start { value of the credit account before the operation, Sum { sum of the operation, Deb n { value of the debit account after the operation, Cred n { value of the credit account after the operation [15].
For reproducing economic dynamics in the model we implement an algorithm
of organizational decision making instead of a production function that is widely
used in macroeconomic models [
The public administration determines structure of the budget, taxation scale, transfer payments, the interest rate and other parameters. Administrative functions are implemented through educational, medical, social security and defense budgetary organizations.
At the current research stage we generate information objects of each module and their standard functions, excluding procedures of decision making of agents and organizations that determine dynamics of the system. 3
Initial modeling data that re ects socio-economic structures at the base year is
presented in Excel tables. The tables contain information on the demographic
structure of the population, organizations and their economic interrelations,
production, import, export, employment, nancial characteristics of organizations
and households, tax rates, transfer payments and other. Information content of
the tables is based on the collections of Federal State Statistics Service [
Information presented in statistical resources requires preprocessing to match ini-tial modeling data structure. For example, to reproduce sector structure of economy in each region we need an aggregated table \Organizations". In the o cial statistics sector structure of the economy, cross-sector interrelations, export and import are presented in the input-output tables; regional production structure is presented in the table \Gross added value of the regions by sectors of the economy" in the statistical yearbook. Direct comparison of data of these two tables is impossible for two reasons. First, calculation of the gross regional product di ers from calculation of the gross domestic product, as a result of which sum of gross regional product in all regions is less than the gross domestic product. Secondly, information on regional production is presented in the form of economic activity types, which implies less detail in comparison with the sector structure (for example, 37 sectors are classi ed as one economic activity type \manufacturing activities"). Thus, regional representation of production in the model requires matching statistical data from di erent sources. Necessary calculations are presented in steps:
1. Calculation of share of each sector in the corresponding type of economic activity on the basis of the input-output table.
dsa = Vs=Va (1) dsa - share of sector s in economic activity type a, Vs - gross product of sector s, calculated by method of added value, Va - gross product of economic activity type a, calculated by method of added value; sector s belongs to economic activity type a.
2. Correction of the table of output of economic activities in regions, taking into account di erence in the domestic product and total amount of regional products in separate economic activities: ka { correction coe cient of economic activity type a; Va { gross product of economic activity type a; vr;a { product of economic activity type a in region r, k presented in statistical tables; vr;a { corrected product of economic activity type a in region r.
3. Completion of the table of output of sectors in regions:
k vs;r = vr;a ds;a (4) vs;r { product of sector s in region r; sector s belongs to economic activity type a.
Preprocessing of initial data is required for all modules except from
Demographics, since the recent information about composition of households in
each region is re ected in results of All-Russian Population Census [
Reconstruction of an arti cial society is carried out in the base year of modeling.
The rst step is to set geographical structure of the Russian Federation (see Fig.
2). After that the original generation of agents is created, distributed among
households and resettled by regions; the composition of households is determined
by data of All-Russian Population Census of 2010 [
Organizations in the model are aggregated: one organization in the model
re-sponds to a set of organizations of one economic sector in the region.
Generation of organizations is based on the input-output table, which determines the
gross output of each economic sector, and the regional distribution of production
table [
Educational institutions are associated with sets of educational places for various groups of specialties and levels of education: school, secondary professional educa-tion; bachelor's, master's or postgraduate courses. Agents of the appropriate age are assigned to educational places.
The generated arti cial society is stored in a database for later use in a series of scenario calculations. 5
Model of an arti cial society was programmed on C# in Microsoft Visual Studio 2015, which is free available for scienti c research. Figure 3 shows the sequence of data processing in the model. The initial modeling data is loaded in the form of Excel tables, after that it is checked for completeness and consistency. In the module of an arti cial society generation, the initial modeling data is transformed to information objects of the model (agents, households and organizations).
Results of the generation procedure are stored in the model database, access to them is provided by SQL-queries. The main resulting tables are accounting entries of organizations, households and state administration. Thus, GDP is calculated as sum of pro ts, wages and taxes paid by commercial, nancial and budget organizations (we use method of calculating GDP by added value). Using grouping queries, it is possible to present gross output in the context of sectors and regions; ows of imports and exports in di erent sectors; household incomes and expenditures; savings and credits of households and organizations.
Interface of the software application of the model is presented in Figure 4 (\Start Modeling" button launches algorithms of dynamics that are being developed at the moment).
Veri cation of the arti cial society model was carried out in two stages. The rst step is validation of algorithms through the test data set. At this stage, the procedures for entering, conversion, storage, retrieval of data and algorithms for generating model objects were checked. At the second stage, the model was veri ed on the basis of retrospective data. For veri cation of the model on retrospective data, we propose a method, which compares demographic, economic and social indicators of the simulation output data with the real values of these indicators (DES-analysis).
The verifying simulation was carried on a time period of one year (12 clock ticks). Population of the Russian Federation was represented by 1.5 million agents, that is, one agent in the model corresponds to 100 residents; for the convenience of further interpretation of the output data, at the end of modeling inverse scaling was performed. Taking into account the accepted assumption that one organization in the model responds to a set of organizations of one economic sector in the region, about 4.5 thousand organizations were created in the model (58 sectors in input-output tables and 90 regions, some sectors are not presented in certain regions). We conducted a series of 10 experiments; the following Table 3 gives a summary of veri cation results.
The set of demographic indicators for comparison includes population
number by sex and age groups, number and composition of households. As
economic indicators we have chosen GDP, gross output, imports and exports of
economic sectors. Social indicators are the average wage, number of unemployed
and number of people below the poverty line. Variation of the observed
parameters is determined by stochastic elements of generation procedures, including
age of agents, composition of households, deviation of wage values in di erent
economic sectors and regions from the average value. To analyze the adequacy
of the model, the output of the generation procedures was averaged over 10 runs
[
Statistical analysis of modeling results for the base year (2014) showed deviation of the selected parameters within 1%, excluding GDP, which variation is about 3%. It is connected mostly with simplifying the tax system in the model: for organizations we set 13% rate, while some sectors of the economy pay additional taxes and it makes signi cant contribution in GDP. Taking this assumption into consideration, we can conclude that calculated variation of modeling parameters indicates a su cient accuracy of arti cial society reconstruction for the base year. However, we should take into consideration that models performing well on data sets available at the time of their publication might perform less well or badly when applied to post-publication data [17]. Due to this reason we plan to verify the model at a few input datasets (years 2015-2017) after they become available in o cial statistics.
For a short period of one year, we estimate deviation of modeling results from retrospective values, which is su cient for assessing similarity of real society and reproduced arti cial society for the base year of modeling. Within calibration of the model on longer time series (at least 5 years), it would be possible to determine autocorrelation and mutual in uence of various factors using methods of regression analysis. 7
The computer model of an arti cial society is designed for reproducing geographical distribution of population of the Russian Federation and its current socio-economic situation. In this paper we describe structure of the initial parameters of the simulation and the procedure for their veri cation. Since the aim of our research is forecasting economic development of Russia and assessing the impact of the state economic policy on this process in our future work we are going to add algorithms of economic dynamics to the model and calibrate them on retrospective data for 2014-2017 period. After o cial registration of the program model, access to the source code would be available through Team Foundation Server of Microsoft Visual Studio 2015. However, to reproduce presented computational results, preprocessed initial data sets should be entered to the model. In order to increase the accuracy of the prognosis we plan to specify the initial modeling data by adding sociological surveys and results of social networks monitoring, which would help to re ect subjective parameters and social moods that are not obvious within standard statistical methods.
Acknowledgments. The reported study was funded by RFBR according to the research project 18-310-00185. 8. Epstein, J., Axtell, R.: Growing Arti cial Societies: Social Science From the Bottom
Up. MIT Press, Brookings Institution (1996) 9. Fonoberova, M., Fonoberov, V.A., Mezi, I.: Global sensitivity/uncertainty analysis for agent-based models. Reliability Engineering & System Safety 118, 8 { 17 (2013). https://doi.org/10.1016/j.ress.2013.04.004, http://www.sciencedirect.com/science/article/pii/S0951832013000999 10. Lee, J.S., Filatova, T., Ligmann-Zielinska, A., Hassani-Mahmooei, B., Stonedahl, F., Lorscheid, I., Voinov, A., Polhill, G., Sun, Z., Parker, D.: The complexities of agent-based modeling output analysis. JASSS : the journal of arti cial societies and social simulation 18(4) (2015) 11. Macy, M.W., Willer, R.: From factors to factors: Computational sociology and agent-based modelling. Annual Review of Sociology 28, 143{166 (2002) 12. Makarov, V.L., Bakhtizin, A.R., Sushko, E.D., Vasenin, V.A., Borisov, V.A., Roganov, V.A.: Supercomputer technologies in social sciences: agent-oriented demographic models. Herald of the Russian Academy of Sciences 86(3), 248{257 (2016) 13. Mashkova, A.L., Demidov, A.V., Savina, O.A., Koskin, A.V., Mashkov, E.A.: Developing a complex model of experimental economy based on agent approach and open government data in distributed information-computational environment. In: eGose '17: Proceedings of the Internationsl Conference on Electronic Governance and Open Society: Challenges in Eurasia. pp. 27{31. ACM international conference proceedings series, ACM, New York, NY, USA (2017) 14. Mashkova, A.L., Novikova, E.V., Savina, O.A.: Agent model for evaluating in uence of tax policy on political preferences. In: EGOSE '16: Proceedings of the International Conference on Electronic Governance and Open Society: Challenges in Eurasia. pp. 258{261. ACM, New York, USA (2016) 15. Mashkova, A.L., Savina, O.A.: Management of nancial ows of organizations in the agent model of the experimental economy. Upravlencheskiy uchet 12, 89{98 (2015), (in Russian) 16. Mashkova, A.L., Savina, O.A., Lazarev, S.A.: Agent model for evaluating e ciency of socially oriented federal programs. In: 11th IEEE International Conference on Application of Information and Communication Technologies (AICT). vol. 2, pp. 217{221. Institute of Control Sciences of Russian Academy of Sciences, Moscow (2017) 17. Moss, S.: Alternative approaches to the empirical validation of agent-based models.
Journal of Arti cial Societies and Social Simulation 11(1), 5 (2008) 18. Ogibayashi, S., Takashima, K.: In uence of the corporation tax rate on gdp in an agent-based arti cial economic system. In: Chen, S., Terano, T., Yamamoto, R., Tai, C. (eds.) Advances in Computational Social Science. Agent-Based Social Systems, vol. 11, pp. 147{161. Springer, Tokyo (2014) 19. Tesfatsion, L.: Agent-based computational economics: Growing economies from the bottom up. Arti cial Life 8(1), 55{82 (2002)