📉 Financial Crises, Exchange Rate Volatility, and Socioeconomic Impact

A Machine Learning Framework for Early Warning Systems in Global Financial Stability
Authors: Rohith Reddy Jonnalagadda, Allu Sagarika, Valluru Anjula Chakravarthy
Affiliation: University of North Texas
Advisor: Dr. Sam Shamroukh

---

📌 Abstract

This project investigates the early detection of financial crises — banking, currency, and systemic — using machine learning and macroeconomic indicators. We utilize a merged dataset combining the Reinhart-Rogoff Global Crisis Dataset and IMF macroeconomic indicators for 70 countries between 1980–2016.

Our best-performing model, XGBoost, demonstrated superior accuracy in predicting systemic and inflation-related crises. The project underscores the utility of AI tools in macroeconomic forecasting and crisis prevention, offering potential support for global financial policy decisions.

---

🎯 Project Objectives

• Identify macroeconomic signals preceding financial crises.
• Evaluate the impact of USD exchange rate volatility on global stability.
• Develop and compare multi-output classification models.
• Recommend machine learning methods for real-time financial crisis monitoring.

---

📁 Current Repository Structure

├── financial_crisis_analysis.ipynb    # Main notebook containing EDA, feature engineering, modeling, and results()
└── README.md                          # Project documentation

Additional scripts and datasets used during development are not included in this public version.

---

🧠 Methodology Summary

• Data Sources:

  • Reinhart-Rogoff Crisis Data (1800–Present)
  • IMF World Economic Outlook Data (1980–2016)

• Data Preprocessing:

  • Missing data handled via MICE (Multiple Imputation by Chained Equations)
  • Normality tested with Shapiro-Wilk
  • Multicollinearity checked via VIF
  • Non-parametric tree-based models chosen due to non-linearity

• Modeling:

  • Multi-output classifiers: XGBoost, Random Forest, K-Nearest Neighbors (KNN)
  • Evaluation using Accuracy, Precision, Recall, F1-Score

---

📊 Key Results

Model	Banking Crisis	Currency Crisis	Systemic Crisis
XGBoost	86% accuracy	86% accuracy	91% accuracy
Random Forest	83% accuracy	87% accuracy	89% accuracy
KNN	78% accuracy	81% accuracy	86% accuracy

Inflation crises were excluded from final conclusions due to perfect accuracy (100%) indicating likely data leakage or overfitting.

---

📌 Key Insights

• Exchange rate volatility and sovereign debt exposure are crucial leading indicators.
• USD appreciation often triggers financial tightening and crisis vulnerability.
• Crises are not exclusive to developing economies — developed countries are also at risk.
• Ensemble models outperform distance-based classifiers in handling macroeconomic data.

---

🧰 Tech Stack

Tool/Library	Purpose
Python	Core programming language
Pandas, NumPy	Data handling
SciPy, Scikit-learn	Statistics, ML models
XGBoost	Gradient boosting classification
Matplotlib, Seaborn	Visualization
R (Tidyverse)	Supplementary statistical tests

---

❓ Research Questions

• Can macroeconomic variables serve as early-warning indicators?
• How do exchange rate and inflation behave prior to crises?
• What is the role of USD dominance in global instability?
• Are ensemble models effective in forecasting multiple crisis types?

---

🔮 Future Directions

• Use higher-frequency data (monthly/quarterly) to improve sensitivity.
• Introduce time-aware models (LSTM, Transformer).
• Incorporate market sentiment/news-based data.
• Cluster countries based on economic characteristics for tailored models.

---

📄 Citation

@unpublished{chakravarthy2025crisisML,
  title={Financial Crises, Exchange Rate Volatility, and Socioeconomic Impact},
  author={Valluru Anjula Chakravarthy, Rohith Reddy Jonnalagadda, Allu Sagarika},
  year={2025},
  institution={University of North Texas}
}

---

🙏 Acknowledgements

Special thanks to Dr. Sam Shamroukh for mentorship, and to peers Raahul Raj Akula and Bhavana Raj Rayapudi for their support during this research.