Quarter Sales Forecasting in Luxury Car Market

The management of the pre-owned vehicle market is a critical strategic pillar for luxury automotive manufacturers, requiring precise sales forecasting to optimize stock sourcing and maximize profit. Currently, the industry relies on a straightforward key performance indicator known as Time-To-Resell (TTR), which estimates expected sales times based on a simple six-month historical average. However, TTR is susceptible to distortions caused by seasonality and vehicle life-cycle stages; the KPI can misleadingly interpret stock shortages (scarcity) as low market demand during the initial ramp-up phases of a model. To address these limitations, this thesis evaluates the efficacy of advanced Machine Learning (ML) models (Linear Regression, XGBoost, and CatBoost) in predicting quarterly vehicle sales at both the macro (MARKET) and micro (DEALER) levels. Utilizing a proprietary dataset from the Ferrari Pre-Owned division across 4 key markets and models, the research establishes a robust forecasting framework that operates as a hybrid between auto-regression and classification. Furthermore, Explainable AI techniques are integrated to ensure the predictive outputs became transparent and interpretable for business stakeholders, successfully identifying stock availability and overall volumes as primary drivers of sales. The experimental results demonstrate that while the traditional TTR method performs adequately during the vehicle’s life-cycle, ML models significantly outperform this state-of-the-art baseline during the complex "tails" of the life cycle. This performance improvement is particularly pronounced at the highly granular (DEALER), where ML successfully captures market dynamics. Ultimately, this thesis advocates for a phased business implementation of these predictive models, starting at the MARKET-Level to build trust before moving to the DEALER-Level, in order to support strategic decision-making and drastically enhance dealership sourcing strategies.