How GlassBox Predictions Work
Complete transparency into our gap momentum methodology. No black boxes.
TL;DR - The Short Version
Strategy: Gap momentum continuation with volume confirmation
Entry signal: Small overnight gaps (0.5-3%) with 1.5x+ volume
Hold period: 3 trading days, then exit at close
Direction: Long only (we only trade gap ups)
ML enhancement: XGBoost predicts gap continuation probability
Backtest results: 59% win rate, +62% return over 5 years
What is Gap Momentum?
A gap occurs when a stock opens at a significantly different price than its previous close. This happens due to overnight news, earnings, or pre-market activity.
Gap Momentum Hypothesis:
Small gaps (0.5-3%) with high volume often continue in the gap direction over the following days. Large gaps (>3%) tend to reverse or "fill", while tiny gaps (<0.5%) lack significance.
Gap Up (Our Signal)
Stock opens above previous close. With volume confirmation, momentum often continues upward.
Gap Down (We Avoid)
Our backtest showed shorting gaps underperforms. We only trade gap ups (long positions).
Entry Criteria (All Must Be Met)
We use strict entry criteria to filter for high-probability setups:
Gap Size
Small gaps continue. Large gaps reverse.
Volume Ratio
Above 20-day average volume confirms conviction.
Direction
Only gap ups. Shorting gaps showed negative edge.
Why These Thresholds?
Through 5-year backtesting across S&P 500 stocks, we found small gap ups (0.5-3%) with volume confirmation had a 67% win rate, while large gaps (>3%) had only 51% win rate (essentially random).
Position Management
Entry
Buy at market open when gap is detected and volume confirms.
Hold Period
3 trading days. Momentum effects dissipate after this window.
Exit
Sell at market close on day 3. No discretionary stops.
Position Sizing
Equal weighting across signals, capped at 10 concurrent positions. Each position = 10% of portfolio. This diversification reduces single-stock risk.
ML Enhancement Layer
While the base gap momentum strategy is profitable, we add an XGBoost classifier to predict which gaps have the highest probability of continuation.
ML Features
- • Gap size (normalized)
- • Volume ratio vs 20-day average
- • RSI (14-day) - momentum context
- • 5-day momentum
- • 20-day trend direction
- • Recent gap success rate (last 5 gaps)
- • Volatility ratio
- • Market regime (SPY trend)
How It Works
- 1. Base strategy identifies gap + volume signal
- 2. ML model scores the setup (0-100% continuation probability)
- 3. We rank all signals by probability
- 4. Select top 10 highest-probability gaps for the week
Backtest Performance (5 Years)
Walk-forward backtest from January 2019 to January 2024 across S&P 500 stocks:
Win Rate
Total Return
Alpha vs S&P
Sharpe Ratio
Important Context
- • These are backtest results, not live trading results
- • Past performance does not guarantee future results
- • Real trading involves slippage, fees, and execution risk
- • Backtest period (2019-2024) included a historic bull market
Gap Type Analysis
Our backtest analyzed different gap sizes and found clear patterns:
| Gap Type | Win Rate | Avg Return | Our Action |
|---|---|---|---|
| Small Gap Up (0.5-3%) | 67% | +0.8% | ✓ Trade Long |
| Large Gap Up (>3%) | 51% | +0.1% | ✗ Skip |
| Small Gap Down | 48% | -0.3% | ✗ Skip |
| Large Gap Down | 45% | -0.5% | ✗ Skip |
This data-driven approach led us to focus exclusively on small gap ups with volume confirmation.
Validation Methodology
We use walk-forward validation to ensure our results are realistic:
Walk-Forward Process
- 1. Train ML model on first 3 years of data (2019-2021)
- 2. Test on next month (January 2022) - record results
- 3. Roll forward: add January 2022 to training set
- 4. Test on February 2022 - record results
- 5. Continue rolling through entire test period
This ensures the model never sees future data during training - the same conditions as making real predictions.
No Data Leakage
- • Features calculated using only past data
- • No look-ahead bias in any indicator
- • Out-of-sample testing on unseen dates
- • Transaction costs included in backtest
Limitations & What We Cannot Predict
- Breaking news: Earnings surprises, FDA decisions, executive departures
- Black swan events: Market crashes, pandemics, geopolitical crises
- Regime changes: The strategy may underperform in bear markets
- Social media pumps: Coordinated retail buying (meme stocks)
Disclaimer: Past performance does not guarantee future results. This is not investment advice. Always do your own research and consult a financial advisor.