Prediction Accuracy
Transparent performance metrics for our sneaker price prediction model. Updated daily.
How Our Predictions Work
SneakerPing uses a machine learning model trained on historical sneaker resale data to predict future price movements. Every day, we evaluate past predictions against actual market prices to measure model performance and calibrate our confidence bands.
What the Metrics Mean
- Direction Accuracy: How often the model correctly predicts whether a sneaker's price will go up or down. Above 50% means the model is better than random chance.
- Average Error: The mean absolute percentage error between predicted and actual prices. Lower is better.
- Within Confidence: How often the actual price falls within the model's confidence band. We target 68% (one standard deviation).
- Sample Size: The number of predictions that have been evaluated. More samples means more reliable metrics.
Prediction Horizons
We generate predictions for three time horizons: 7-day (short-term flips), 30-day (monthly trends), and 90-day (seasonal movements). Short-term predictions tend to have higher direction accuracy but smaller absolute price changes, while longer horizons capture bigger moves with more uncertainty.
Continuous Improvement
Our model runs a daily feedback loop that compares heuristic and ML predictions, adjusts the ensemble weight between them, and recalibrates confidence bands based on observed performance. This page shows the results of that process in real time.