The article focuses on developing and testing a methodology for quantitatively assessing the robustness of time series models (SARIMA, Prophet, LSTM) used in inventory management under unstable external conditions. The proposed approach involves modeling input data distortion scenarios, including holiday demand surges, logistical disruptions, inflationary trends, and structural shifts. Robustness metrics—Robustness Index (RI) and Degradation Ratio (D)—are introduced to evaluate forecast degradation. Experiments on synthetic data show that high accuracy on clean data does not guarantee robustness: SARIMA is sensitive to inflationary trends, Prophet is robust to seasonality, and LSTM is vulnerable to structural shifts. The findings are applicable in logistics and retail for optimizing supply planning.

Keywords: inventory management, time series, model robustness, demand forecasting, SARIMA, Prophet, LSTM, Robustness Index, Degradation Ratio, logistical disruptions, seasonal fluctuations

`

Keywords: