A Universal Model for Debt Transparency Based on The Forecast of The Ratio of Debt to GDP

Debt transparency is a critical global issue, especially for countries that face difficulties in reporting public debt. The aim of this study is to develop a universal prediction model, especially for countries with data deficiency and reliability issues, by using various machine learning methods such as Support Vector Regression (SVR), XGBoost, Random Forest, Ridge Regression, and Lasso Regression to estimate the debt-to-GDP ratio of PIIGS countries. In the study, the debt-to-GDP ratios of PIIGS countries were estimated using quarterly data from 2000:Q1 to 2021:Q4 in five different experiments, along with robustness tests. In the first experiment, the consistency of each country's own data was evaluated and the models resulted in high accuracy. In the second experiment, a universal model was created with only the data of other countries, but the success of the models decreased. In the third, fourth, and fifth experiments, the success rate was improved by adding the data of the test countries to the model. In particular, universal models created with Random Forest and XGBoost predicted the debt-to-GDP ratio in the test countries with high accuracy, with R² values reaching up to 99.9% in Experiment 5. This study contributes to debt transparency by providing a reliable model to estimate debt-to-GDP ratios for countries with limited data. It also emphasizes the importance of the method of making predictions by using data from similar countries for countries with missing or low-reliability data. In this study, complex machine learning models such as Random Forest and XGBoost achieve the highest success in debt-to-GDP ratio prediction. The study helps international institutions improve debt management and transparency.