The article presents a comprehensive methodology for forecasting macroeconomic indicators for long-term planning in project finance. The purpose of the research is to develop a system of stochastic simulations capable of generating plausible scenarios of economic development, taking into account the relationships between various economic parameters. The methodology includes two key components: an algorithm for selecting significant predictors based on sparse graphs and the minimum Steiner tree, and a system of stochastic simulations integrating the CIR++ model with the Monte Carlo method. The author has developed an efficient algorithm for building regression models that takes into account structural relationships between economic indicators. The research material consisted of historical data on a wide range of Russia's macroeconomic indicators: GDP, inflation, interest rates, real estate price indices, and loan delinquency rates. The results of applying the methodology demonstrate high accuracy of forecasting on historical data and intuitively understandable behavior in the long term. Model validation is based on conceptual validity, systematic output analysis, and business logic verification rather than traditional point forecast metrics, which is appropriate for long-term scenario generation. PCST hyperparameter calibration methodology and extreme scenario modeling for tail risk assessment are presented. The system is capable of generating probabilistic scenarios with a horizon of up to 30 years, which allows assessing various aspects of risks, including extreme scenarios. The modular architecture of the system provides flexibility and adaptability to various economic conditions. The results of the research have practical significance for risk management in financial institutions and strategic planning in project finance.