Improving statistical obsolescence forecasting accuracy in the absence of covariate predictors through time-aware validation and binning strategy

Obsolescence management is pivotal in ensuring a continuous supply of products and services within a supply chain. Accurately forecasting the probability and timing of obsolescence is crucial for companies across the product lifecycle, from design and procurement to supplier selection. This issue is particularly pronounced in the defense industry, where systems are often procured in low volumes, tailored to specific requirements, and purchased irregularly. This study evaluates the accuracy of statistical methods for forecasting product obsolescence in scenarios with limited covariate predictors, using an operational data set. Results indicate that static binning offers more accurate forecasts at the part level. In contrast, dynamic binning is more robust against extreme forecast errors, particularly for larger bin sizes and when lifecycle forecasting is used. Moreover, bin size is a statistically significant factor in forecast accuracy, with greater accuracy at smaller bin sizes. However, when forecasting obsolescence risk, larger bin sizes are more advantageous for parts further along in their lifecycles, with performance further influenced by the probability of obsolescence. These findings have practical implications and provide a foundation for more reliable obsolescence predictions, enabling informed decisions regarding reactive, proactive, and strategic obsolescence management.