Forecasting Kenya's public debt using time series analysis

Abstract

Accurately forecasting public debt is essential for developing countries like Kenya to maintain fiscal sustainability and economic stability. This study aimed to identify the best time series forecasting model for predicting Kenya's future public debt to help policymakers create effective fiscal reforms. The Autoregressive Integrated Moving Average (ARIMA) and Holt-Winters exponential smoothing models were tested due to their ability to handle complex patterns and seasonality in time series data. Public debt data from Kenya from 2001 to 2021 were analyzed, and both models were applied to the processed data. The ARIMA (0,2,1) model, which uses second-order differencing and a moving average component, was found to be the best model based on information criteria. The Holt-Winters additive method also showed good performance, adapting well to recent data and seasonal trends with optimized smoothing parameters. Both models produced forecasts that closely matched the actual debt figures for 2022 and 2023, with an error margin of only 0.73. Measures of accuracy, such as Mean Absolute Percentage Error (MAPE) and Mean Absolute Scaled Error (MASE), confirmed the reliability of the models, with ARIMA performing slightly better than Holt-Winters. While previous studies have looked at debt forecasting for Kenya, this research offers a thorough evaluation and comparison of two strong time series models. Unlike existing literature, this study provides a rigorous out-of-sample forecasting assessment, identifying the best approach for reliably predicting Kenya's debt. However, the study is limited by its focus on univariate time series models, which could be improved by including relevant external economic variables. The findings show that the ARIMA and Holt-Winters models are accurate tools for forecasting Kenya's public debt, helping policymakers to develop sustainable debt management strategies and fiscal reforms based on reliable future projections.