Abstract

This study aims to enhance the accuracy of GDP growth forecasting in Sri Lanka by evaluating structural time series decomposition models, with a focus on Harvey-type models that incorporate intervention analysis. Despite the importance of reliable forecasts for economic planning, limited studies have applied these models to GDP data in the Sri Lankan context. Addressing this gap, the study utilises quarterly GDP growth rates from 2000 to 2023, sourced from annual reports of the Central Bank of Sri Lanka. A range of Harvey-type structural time series specifications were examined, with model performance assessed using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Akaike’s Information Criterion (AIC), Bias-corrected AIC (AICc), and Bayesian Information Criterion (BIC). The chosen model incorporates time-varying level and trend along with intervention terms to account for external shocks. Forecasts for the four quarters of 2024 closely matched actual values, with predicted GDP growth rates of 5.279%, 4.652%, 5.445%, and 5.128%, compared to actuals of 5.3%, 4.7%, 5.5%, and 5.3%, respectively. The results confirm the model's robustness and the resurgence of the agriculture, industry, and service sectors in Sri Lanka’s post-crisis economy. This research highlights the flexibility and predictive strength of Harvey-type structural time series models with intervention analysis. It recommends their adoption for economic forecasting in Sri Lanka, particularly under conditions of volatility, thereby offering a valuable tool for policymakers and stakeholders engaged in forward-looking economic planning.

Keywords: Dynamic forecasting, GDP growth, Harvey-type models, Intervention analysis, Time series decomposition