Structural Dynamics and Predictive Performance of Agricultural Production in Sub-Saharan Africa: A Robust Forecasting Approach Under Data Quality Constraints

Agricultural production forecasting in Sub-Saharan Africa faces persistent challenges: data quality constraints, structural breaks from conflict and policy shifts, and methodological fragmentation. This study develops and validates a methodological framework specifically designed for data-constrained environments. Using national agricultural production data (1961-2024) for five Southern African countries and six crops, we evaluate three dimensions: (1) rigorous quality criteria (minimum 20 observations, maximum 3-year gaps) for series inclusion; (2) integration of Bai-Perron structural break detection with historical contextualization; and (3) comparative assessment using Diebold-Mariano tests. Of 26 potential series, 19 met quality criteria (MAPE ≤ 25%), yielding a mean MAPE of 12.5% (range: 1.6-24.5%). Models incorporating structural breaks achieved MAPE reductions of 17-23% compared to specifications assuming parameter stability. XGBoost outperformed ARIMA for volatile crops (maize), while ARIMA performed better for stable crops (beans, cassava), confirmed by Diebold-Mariano tests (p < 0.05 for 4 approved series). Projections reveal heterogeneous patterns: Malawi’s soybean production is projected to decline (-7.2%), while Angola’s soybean production is projected to remain relatively stable (-0.0%). These patterns coincide with documented structural conditions, though causality is not directly tested. These findings indicate that in data-constrained environments, forecast reliability depends as much on data quality control and structural break integration as on model sophistication. The framework offers a replicable methodology for agricultural forecasting under data constraints, with implications for food security planning across developing regions.