Evaluating Inventory Forecasting Under Demand Shocks in Automotive Parts Suppliers

Thesis Abstract

The automotive parts supply chain faces recurrent demand shocks driven by macroeconomic volatility, geopolitical disturbances, and disruptions in production networks, which exacerbate forecasting errors, inflate inventory carrying costs, and increase stockouts. This study addresses how inventory forecasting performs under demand shocks and seeks to identify robust forecasting practices that sustain service levels while minimizing total cost in automotive parts suppliers. The aim is to develop and validate a forecasting framework that adapts to irregular demand patterns and to quantify the resilience of inventory policies under shock conditions. Specific objectives include (i) characterizing the nature, frequency, and magnitude of demand shocks experienced by automotive parts suppliers over a ten-year window, (ii) evaluating the performance of conventional forecasting methods (time-series, causal, and machine learning models) against shock-aware approaches using historical and simulated shock scenarios, (iii) assessing the impact of forecast accuracy on inventory metrics such as fill rate, stock range, days of inventory on hand, and total cost of ownership, (iv) examining the moderating role of supplier collaboration and information sharing on forecast error propagation, and (v) proposing a practical, implementable inventory forecasting framework tailored to automotive parts suppliers operating in high-variability environments. The study adopts a quantitative, mixed-methods research design combining retrospective data analysis with scenario-based simulations. The population encompasses Tier-1 and Tier-2 automotive parts suppliers in North America and Europe, with a stratified sample of 30 firms representing diverse product families (engine components, electrical systems, braking systems) and varying demand variability profiles. Within each firm, monthly historical data for the past ten years will be compiled, including forecast figures, actual demand, inventory levels, service levels, lead times, procurement costs, and supply disruption indicators. In addition, semi-structured interviews with supply planners (n=45) will inform the qualitative dimension of the forecasting process and policy levers. Data collection instruments include standardized data extraction templates, forecast error logs, and interview guides validated through expert review. Quantitative analysis will proceed in three stages. First, descriptive statistics will profile demand shock characteristics and baseline forecast performance across methods. Second, forecast accuracy will be evaluated using metrics such as mean absolute percentage error (MAPE), weighted MAPE, and directional accuracy under both normal and shock conditions, with shock episodes identified via anomaly detection algorithms (e.g., CUSUM and EWMA) and corroborated by disruption indices. Third, predictive performance will be compared across models—traditional ARIMA, exponential smoothing, Prophet, and machine learning approaches (random forest, gradient boosting, and long short-term memory networks)—under augmented datasets that incorporate shock indicators (volatility indices, disruption dummies, and exogenous covariates). A hierarchical Bayesian approach will also be employed to quantify uncertainty and to fuse forecasts across product families. Inventory policy implications will be assessed through simulation modeling (discrete-event simulation and system dynamics) to estimate expected fill rate, service level, and total cost under varying shock severities and recovery speeds. The study will also test the moderating effects of information-sharing practices and supplier collaboration using interaction terms in regression analyses and structural equation modeling to elucidate causal pathways. Expected findings include (i) superior resilience of shock-aware forecasting models—particularly those incorporating disruption indicators and exogenous covariates—over traditional methods in maintaining service levels while reducing excess inventory, (ii) quantification of the cost trade-offs between forecast accuracy and inventory carrying costs under shocks, and (iii) evidence that enhanced information sharing with suppliers and customers mitigates forecast error amplification and stabilizes inventory positions. The study contributes to knowledge by integrating shock dynamics into inventory forecasting theory, augmenting operations management literature with empirically validated, industry-specific models for automotive parts, and providing a replicable framework for practitioners to implement forecast-informed inventory policies under uncertainty. The main conclusion is that incorporating explicit shock signals and collaborative information flows into forecasting processes yields meaningful improvements in both service levels and total cost performance. Practical recommendations include adopting hybrid forecasting ensembles with shock-aware features, investing in data integration across the supply network, and establishing formal information-sharing protocols with suppliers to strengthen demand visibility and response agility.