SneakerPulse Backtest Engine — Quentin's Method

How This Prediction Works

Our model uses a volatility-anchored approach: the target shoe's actual current price is the starting point, and comparable shoes inform the trajectory and uncertainty.

1. Data Collection: We pull real sales data from GOAT for the target shoe and 8 comparable Air Jordan 5 colorways. Each sale includes timestamp, price, and size.
2. Current Price Anchoring: Predictions start from the shoe's actual current resell price ($373), derived from 1062 pre-release transactions at avg $362, adjusted by the pre-to-release discount factor. Comparables inform trajectory shape and volatility, not price levels.
3. Size Outlier Filtering: Extreme prices from rare sizes (GS, 17+) are removed using IQR-based filtering (2× IQR beyond Q1/Q3). This prevents outlier sales from skewing the predicted max or confidence intervals.
4. Pre-Release Ratio Weighting: Comparables weighted by how closely their pre-release price-to-retail ratio matches the target's 1.65x ratio. Exponential decay: w = e^{-3.0·|r_comp-r_target|} × volume bonus. Steeper penalty ensures only similarly-priced shoes heavily influence predictions.
5. Trajectory Shape Extraction: Each comparable's post-release price curve is normalized to its own Day-0 price, producing a shape (e.g., 'drops 15% by Day 30'). The weighted composite shape is applied to the target's actual price.
6. Monte Carlo Simulation: 10000 simulations perturb the trajectory shape (±12%) and add volatility noise (weighted σ=0.210) to generate confidence intervals.
7. Pre-Release Discount: Comparables show -2.9% avg change from pre-release to release day. Applied as a smoothing factor that fades over 14 days.
8. Release-Day Intraday: 15-min buckets for first 24h, 30-min for next 24h from 8 comparable release patterns, with 5000 MC sims for uncertainty.
9. Size Premiums: Degree-3 polynomial regression on size-price data, averaged across comparables, captures the 'size curve' (e.g., size 14 premium, size 7 discount).
10. Risk Metrics: VaR 95% = worst return at the 5th percentile. Expected Shortfall (ES) = average return below VaR. Both computed from the MC distribution of returns relative to retail.