A GAME-THEORETIC MODEL OF THE INTERACTION OF E-COMMERCE ENTERPRISES UNDER NETWORK EXTERNALITIES

Abstract

The article presents a comprehensive model of the game-theoretic interaction of enterprises in the e-commerce market. The model functions under conditions of network externalities. The relevance of the study is due to the saturation of the e-commerce market with IT solutions and many interconnected participants. This requires the development of flexible and realistic tools for cybernetic modeling and competitiveness analysis. The main element of the model is the integration of game theory and machine learning (ML) methods. Unlike common classical models based on the assumption of perfect rationality of players, the study applies the concept of bounded rationality through the mechanism of probabilistic strategies of Quantal Response Equilibrium (QRE). The proposed approach made it possible to adequately model the behavior of e-commerce companies, since these companies strive for profit but make decisions under conditions of incomplete information and limited resources. These constraints a priori make their actions not always optimal. Each enterprise is considered as an agent. The agent in the model chooses a strategy vector that includes price, quality, marketing intensity, and investment in R&D. The scientific novelty of the model lies in the formalization and combination of several key parameters. First, an integrated competitiveness index (), was developed, which, unlike classical financial metrics, takes into account the synergy of four factors: economic result (expected profit), the structural significance of the enterprise in the interaction network (criticality weight), the robustness of its strategy to market disturbances, and innovation potential. The latter was evaluated by the dynamics of R&D investment. Second, the equilibrium state in the system is described not by the classical Nash equilibrium but through the more general apparatus of variational inequalities. This made it possible to correctly consider convex cost functions and regularization elements. Third, the application of machine learning algorithms made it possible to calibrate the model parameters based on empirical data. This turns the model into a tool for forecasting and managerial analysis of companies in the e-commerce market. A training algorithm for the model was developed and presented in the form of a flowchart and pseudocode.