Methodology

How we collect, classify, and analyze high-end, classic, and enthusiast car market data

Data Sources

Hypercars aggregates vehicle listings from 100+ sources including Bring a Trailer, Cars & Bids, eBay Motors, RM Sotheby's, Mecum, PCarMarket, Collecting Cars, and enthusiast forums like Rennlist. This includes both auctions and fixed-price/classified listings. Data is indexed continuously throughout the day and processed through our classification pipeline.

AI Classification

Every listing is classified into a canonical make, model, and submodel using a two-stage pipeline. The first stage uses exact keyword matching against our taxonomy of 500+ makes and thousands of models, achieving 85-90% coverage. Listings that don't match exactly go through a Gemini LLM fallback that handles misspellings, alternate names, and ambiguous titles. Each classification receives a confidence level (HIGH, MEDIUM, or LOW) and is validated against our taxonomy.

Price Predictions

We use an AutoGluon machine learning model trained on historical sales data to predict fair market values. The model considers year, make, model, submodel, mileage, transmission type, and recent comparable sales. Predictions are updated as new transactions occur.

Deal Scoring

Each listing receives a deal ratio calculated as actual sale price divided by predicted price. A deal ratio below 1.0 means the car sold for less than its predicted value — a potential good deal. A ratio above 1.0 means it sold above predicted value. We use this to rank the best and worst deals on the platform.

Appreciation Tracking

We track how models are trending by comparing current predicted prices against predicted prices from one year ago. The appreciation ratio (predicted price now / predicted price 1 year ago) shows which models are gaining or losing value. Ratios above 1.0 indicate appreciation.

Fear & Greed Index

The Hypercars Fear & Greed Index is a composite market-activity and demand indicator. It equally weights average daily sold volume, average daily median bid activity per ended auction, and sell-through during the latest seven days. Each factor is normalized against its mean and standard deviation during the preceding 90 days, clipped to limit outliers, then combined into a 0–100 score. It is not a price forecast or a literal measurement of emotion.

Data Coverage

Hypercars tracks 4,000,000+ listings across 100+ sources, indexing 25,000+ new listings a day. Coverage spans from vintage classics to modern performance cars, including detailed specs like mileage, transmission, color, and VIN where available.