Browsing by Author "Eyasu Desta"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Long-Term Inflation Trend Prediction In Ethiopia Using LSTM And ARIMA Ensemble Model
    (Addis Ababa University, 2025-03) Eyasu Desta; Bisrat Derebssa (PhD)
    Accurate inflation forecasting is crucial for economic stability, influencing policy formulation, financial planning, and market predictions. In Ethiopia, inflation dynamics are shaped by complex, interdependent factors, including macroeconomic indicators and sudden economic shocks such as civil war and drought. Traditional methods like ARIMA excel at capturing linear trends but struggle with non-linearities and external influences. Conversely, machine learning approaches like LSTM neural networks effectively model non-linear dynamics but often require extensive data and complex parameter tuning, limiting their applicability in data-scarce environments. To address these limitations, this study introduces an ensemble forecasting model that integrates ARIMA and LSTM, leveraging the strengths of both methodologies. The ensemble model uses historical economic data from Ethiopia spanning 1979 to 2023, incorporating key features such as money supply, real GDP, government investment, interest rates, exchange rates, and binary event flags representing sudden economic shocks. By combining ARIMA's proficiency in linear trend detection with LSTM's ability to model complex, nonlinear relationships, the ensemble approach achieves superior predictive accuracy. The study evaluates the model's performance under varying conditions, including different historical data lengths for training and testing, ensuring adaptability to diverse data availability scenarios. Additionally, the ensemble model handles multiple prediction horizons, providing reliable forecasts for inflation trends. This flexibility enhances its utility for policymakers and economists. The inclusion of binary event flags for civil war and drought ensures the model accounts for sudden disruptions, which significantly impact inflation in developing economies. Results indicate a substantial improvement in accuracy, with the ensemble model achieving 97.77% accuracy, outperforming standalone ARIMA (78.32%) and LSTM (89.01%). This highlights the model's effectiveness in capturing both linear and nonlinear patterns while addressing external shocks. The findings underscore the ensemble model's potential as a robust tool for economic forecasting and policymaking, facilitating informed fiscal strategies. Future research could explore further customization, integration of additional macroeconomic variables, and enhanced responsiveness to economic shocks, ensuring even greater predictive capability.