How TCGPredict Predicts Card Prices
We built a machine learning model that analyses what makes Pokemon cards valuable. Here's how it works at a high level.
The Approach
We Study the Fundamentals
The model looks at what actually drives card value: how hard it is to pull from packs, how popular the Pokemon is, how rare the artwork is, how old the card is. Not just what someone listed it for on TCGPlayer.
We Compare Against the Market
Every card with a market price gets a predicted fair value based on its fundamentals. When the market price is significantly below our prediction, that's a potential opportunity. When it's above, the card might be overpriced.
We Validate With Real Data
The model was backtested over 14 months using real historical prices. 85% of cards flagged as undervalued actually went up. That's not a simulation, that's what happened.
19,583
Cards Scored
94.5%
Price Variance Explained
85%
Backtest Win Rate
14mo
Historical Validation
Data Sources
Prices from Pokemon TCG API (updated daily)
eBay UK sold prices (real completed sales, not listings)
PSA grading data from population reports
No made-up data. Every number comes from a verified source.
What We Don't Do
We don't guarantee profits
We don't factor in card condition (the model assumes near-mint)
We don't predict short-term spikes from hype or YouTuber openings
We don't use any pricing data as an input to avoid circular logic
Frequently Asked Questions
Ready to find undervalued cards?
Browse cards the model thinks are priced below fair value.
View Undervalued Cards