Kalimati Vegetable Price Index Forecasting with a Momentum Corrected Online Stacking Ensemble

Researchers developed the Kalimati Vegetable Price Index (KVPI), a composite index tracking 135 daily wholesale commodities from Nepal over ten years, using a momentum-corrected ensemble model to forecast agricultural prices with 0.68% error at 90-day horizons. The tool addresses forecasting challenges in emerging markets and provides policymakers with actionable insights for food security planning.

Agricultural price forecasting in developing economies presents unique challenges that traditional models struggle to address. Nepal's Kalimati market, a major wholesale hub, experiences extreme volatility driven by supply disruptions, seasonal demand patterns, and cultural consumption cycles. This study tackles those complexities by constructing a macro-level price index rather than modeling individual crops, effectively filtering noise while capturing systemic market movements.

The research reflects a broader shift toward specialized machine learning applications in emerging market infrastructure. By aggregating 135 commodities into a single signal and engineering 64 causally valid features—including festival timing and rolling statistics—the researchers created a framework that respects local market dynamics rather than applying generic global models. The finding that tree-based ensembles outperformed transformer architectures is noteworthy, suggesting that interpretability and robustness matter more than raw model complexity for volatile, noisy datasets.

The practical implications extend beyond academia. Food price volatility directly impacts inflation, consumer purchasing power, and policy intervention decisions in emerging economies. Accurate 90-day forecasts enable supply chain actors to manage inventories more efficiently and help governments implement timely price stabilization measures. The open-source release multiplies impact by enabling adoption across similar markets in South Asia and beyond.

The momentum-corrected stacking ensemble's performance—84.5% variance explained with 0.68% MAPE—sets a credible benchmark for agricultural forecasting. Future work should explore how such models integrate with climate data and cross-border trade dynamics to enhance predictive power across longer horizons.

• →A momentum-corrected ensemble model forecasts Nepalese vegetable prices with 0.68% error, outperforming complex transformers on noisy agricultural data.
• →Aggregating 135 commodities into a composite index reduces noise while capturing systematic price movements across Nepal's largest wholesale market.
• →Tree-based ensembles proved more reliable than deep learning on volatile datasets with limited historical patterns, challenging conventional ML hierarchies.
• →Accurate price forecasting strengthens food security policy and supply chain optimization in emerging markets with high commodity volatility.
• →Open-source pipeline enables replication across similar markets in South Asia, scaling impact beyond Nepal's Kathmandu market.