Code     

V5-2508

Project

Forecasting public finance trends using artificial intelligence

Period

01.09.2025 – 31.08.2026

Head

Riste Ichev, Ph.D.

Research activity

Social science/ Economics

Abstract

Traditional methods of fiscal forecasting have largely relied on econometric models based on linear relationships between macroeconomicvariables, assuming stable conditions for reliable predictions. However, the growing complexity of the global economic environment and increasingfrequency of shocks (e.g., pandemics, energy crises, inflationary pressures) have revealed the limitations of classical approaches. There is a growingneed for alternative, data-driven methods that can respond more dynamically to real-world conditions. For this challenge, we will use artificialintelligence methods, which were exponentially developed in the last years.

The objective of this project is to develop an experimental artificial intelligence model that enhances the accuracy of fiscal forecasts by analyzinghistorical time series of public finance data. We will employ machine learning (ML) and deep learning (DL) techniques, focusing on Recurrent NeuralNetworks (RNN), Long Short-Term Memory networks (LSTM), and Transformer architectures – they have recently demonstrated state-of-the-artperformance in modeling large datasets, contextualizing information, and identifying patterns. The project will involve training and evaluating variousmodel architectures using historical fiscal and macroeconomic data, with particular emphasis on volatility, seasonality, and input complexity.

A prototype model will be developed in a controlled simulation environment and trained through supervised learning. The project will examine theinterpretability of models, technical implementation requirements in institutional settings, and benchmark them against existing macroeconometricmodels. The goal is to identify the added value of AI in fiscal forecasting, understand where it can complement traditional tools, and assess itscapacity to fiscal planning, and potential evidence-based policymaking. The results will contribute to a deeper understanding of AI’s potential andlimitations in fiscal modelling, set the foundation for further research, and provide guidance for integration into public sector institutions responsiblefor fiscal strategy and economic policy.

Researchers 

Forecasting public finance trends using artificial intelligence                                           

The phases of the project and their realization

The project will be carried out in several stages:

1. Data collection and preparation: this includes obtaining historical time series of fiscal aggregates, macroeconomic indicators (e.g., GDP, inflation, unemployment, interest rates), as well as data on political events and institutional changes.
2. Model development: this will involve designing multiple variations of core models (RNN, LSTM, Transformer), with a focus on selecting appropriate hyperparameters, structuring input sequences, and using different techniques to improve model performance (e.g., dropout).
3. Model training and validation: we will apply a supervised learning approach with separate training and testing datasets. Evaluation will be based on statistical metrics such as RMSE, MAE, MSE, and the coefficient of determination (R²).
4. Comparison with existing models: we will analyze situations in which AI-based models are more accurate, stable, and efficient compared to macro-econometric approaches.
5. Development of implementation guidelines: this will include technical documentation, staffing requirements, and an assessment of the potential impact of deploying such a model within public administration.

Citations for bibliographic records

Forecasting public finance trends using artificial intelligence [V5-2508]