Robust Time Series Forecasting of U.S. Vehicle Sales Under Economic Shocks

Forecasting vehicle sales is essential for production planning, supply-chain management, and market strategy in the automotive industry. However, vehicle sales are strongly influenced by economic cycles and unexpected shocks, making accurate forecasting challenging. This study analyzes U.S. monthly vehicle sales from ۱۹۷۶ to ۲۰۲۲ using official data from the Federal Reserve Economic Database (FRED). Three forecasting strategies are evaluated: Exponential Smoothing with a fixed window, Exponential Smoothing with an expanding window, and Linear Regression with Recursive Feature Elimination and Cross-Validation (RFECV) using an expanding window approach. Model performance is assessed using MAPE, MAE, and distributional error metrics. The results show that expanding-window approaches provide more stable forecasting performance under non-stationary conditions. Major economic disruptions such as the ۲۰۰۸ financial crisis and the COVID-۱۹ pandemic significantly affect forecasting accuracy. The findings highlight the importance of adaptive forecasting strategies for real-world economic time series and provide practical insights for data-driven decision making in the automotive sector.

کلیدواژه ها:

Time series forecasting ، vehicle sales ، economic shocks ، machine learning ، expanding window