WHAT WAS LOST: Estimating censored demand in family planning supply chains with the Truncated Conformal Kalman Filter

Forecasting
Inventory
Loss Sales
R
Tobit Kalman Filter
Conformal Prediction
5th Welsh Postgraduate Research Cluster Workshop in Economy, Enterprise and Productivity
Author

Harsha Halgamuwe Hewage

Published

September 16, 2025


Context

At the 5th Welsh Postgraduate Research Cluster Workshop in Swansea University, we present a new lens on forecasting for public health supply chains, where demand signals are often distorted by stockouts and service interruptions. When shelves are empty, demand doesn’t disappear, it goes unrecorded.

We propose a Truncated Conformal Kalman Filter (TCKF) that reconstructs censored demand while generating uncertainty-aware forecasts, directly usable in inventory planning. Going beyond accuracy metrics, we evaluate how forecasts translate into inventory efficiency and ultimately public health outcomes.

Using synthetic and real-world data from Côte d’Ivoire, we show how better forecasting doesn’t just improve service levels, it prevents stockouts, reduces unmet need, and saves lives. Our work offers a reproducible framework that links forecasting with reorder decisions and health impact—because what gets forecasted, gets funded and delivered.